gistfile1.txt

# 1 "E:/github.com/andlabs/libui/windows/utf16.cpp"
# 1 "E:\\github.com\\andlabs\\libui\\build//"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "E:/github.com/andlabs/libui/windows/utf16.cpp"

# 1 "E:/github.com/andlabs/libui/windows/uipriv_windows.hpp" 1

# 1 "E:/github.com/andlabs/libui/windows/winapi.hpp" 1
# 26 "E:/github.com/andlabs/libui/windows/winapi.hpp"
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 12 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_mac.h" 1 3
# 98 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_mac.h" 3
             
# 107 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_mac.h" 3
             
# 13 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_secapi.h" 1 3
# 44 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_secapi.h" 3

# 44 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_secapi.h" 3
extern "C++" {
template <bool __test, typename __dsttype>
  struct __if_array;
template <typename __dsttype>
  struct __if_array <true, __dsttype> {
    typedef __dsttype __type;
};
}
# 14 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 2 3
# 282 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/vadefs.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/vadefs.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 578 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sdks/_mingw_directx.h" 1 3
# 579 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sdks/_mingw_ddk.h" 1 3
# 580 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 2 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/vadefs.h" 2 3




#pragma pack(push,_CRT_PACKING)



extern "C" {





  typedef __builtin_va_list __gnuc_va_list;






  typedef __gnuc_va_list va_list;
# 99 "C:/msys64/mingw64/x86_64-w64-mingw32/include/vadefs.h" 3
}



#pragma pack(pop)
# 283 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 2 3
# 541 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 3
extern "C" {
# 552 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 3
void __attribute__((__cdecl__)) __debugbreak(void);
extern __inline__ __attribute__((__always_inline__,__gnu_inline__)) void __attribute__((__cdecl__)) __debugbreak(void)
{
  __asm__ __volatile__("int {$}3":);
}




const char *__mingw_get_crt_info (void);


}
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sdkddkver.h" 1 3
# 11 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 2 3
# 65 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 11 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 2 3




#pragma pack(push,_CRT_PACKING)
# 35 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
__extension__ typedef unsigned long long size_t;
# 45 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
__extension__ typedef long long ssize_t;






typedef size_t rsize_t;
# 62 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
__extension__ typedef long long intptr_t;
# 75 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
__extension__ typedef unsigned long long uintptr_t;
# 88 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
__extension__ typedef long long ptrdiff_t;
# 106 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
typedef unsigned short wint_t;
typedef unsigned short wctype_t;





typedef int errno_t;




typedef long __time32_t;




__extension__ typedef long long __time64_t;
# 138 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
typedef __time64_t time_t;
# 422 "C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h" 3
struct threadlocaleinfostruct;
struct threadmbcinfostruct;
typedef struct threadlocaleinfostruct *pthreadlocinfo;
typedef struct threadmbcinfostruct *pthreadmbcinfo;
struct __lc_time_data;

typedef struct localeinfo_struct {
  pthreadlocinfo locinfo;
  pthreadmbcinfo mbcinfo;
} _locale_tstruct,*_locale_t;



typedef struct tagLC_ID {
  unsigned short wLanguage;
  unsigned short wCountry;
  unsigned short wCodePage;
} LC_ID,*LPLC_ID;




typedef struct threadlocaleinfostruct {
  int refcount;
  unsigned int lc_codepage;
  unsigned int lc_collate_cp;
  unsigned long lc_handle[6];
  LC_ID lc_id[6];
  struct {
    char *locale;
    wchar_t *wlocale;
    int *refcount;
    int *wrefcount;
  } lc_category[6];
  int lc_clike;
  int mb_cur_max;
  int *lconv_intl_refcount;
  int *lconv_num_refcount;
  int *lconv_mon_refcount;
  struct lconv *lconv;
  int *ctype1_refcount;
  unsigned short *ctype1;
  const unsigned short *pctype;
  const unsigned char *pclmap;
  const unsigned char *pcumap;
  struct __lc_time_data *lc_time_curr;
} threadlocinfo;







#pragma pack(pop)
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 2 3

#pragma pack(push,_CRT_PACKING)


extern "C" {


  struct _EXCEPTION_POINTERS;
# 43 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 3
  struct _EXCEPTION_RECORD;
  struct _CONTEXT;
# 54 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 3
  unsigned long __attribute__((__cdecl__)) _exception_code(void);
  void *__attribute__((__cdecl__)) _exception_info(void);
  int __attribute__((__cdecl__)) _abnormal_termination(void);






  typedef void (__attribute__((__cdecl__)) * _PHNDLR)(int);

  struct _XCPT_ACTION {
    unsigned long XcptNum;
    int SigNum;
    _PHNDLR XcptAction;
  };

  extern struct _XCPT_ACTION _XcptActTab[];
  extern int _XcptActTabCount;
  extern int _XcptActTabSize;
  extern int _First_FPE_Indx;
  extern int _Num_FPE;

  int __attribute__((__cdecl__)) __CppXcptFilter(unsigned long _ExceptionNum,struct _EXCEPTION_POINTERS * _ExceptionPtr);
  int __attribute__((__cdecl__)) _XcptFilter(unsigned long _ExceptionNum,struct _EXCEPTION_POINTERS * _ExceptionPtr);





  typedef int (*PEXCEPTION_HANDLER)(struct _EXCEPTION_RECORD*, void*, struct _CONTEXT*, void*);
# 125 "C:/msys64/mingw64/x86_64-w64-mingw32/include/excpt.h" 3
}


#pragma pack(pop)
# 66 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 2 3
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/stdarg.h" 1 3 4
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdarg.h" 1 3 4
# 140 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdarg.h" 3 4
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_stdarg.h" 1 3 4
# 141 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdarg.h" 2 3 4
# 2 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/stdarg.h" 2 3 4
# 67 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windows.h" 2 3


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/windef.h" 1 3







# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 1 3



# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 5 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winapifamily.h" 1 3
# 6 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/specstrings.h" 1 3
# 12 "C:/msys64/mingw64/x86_64-w64-mingw32/include/specstrings.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sal.h" 1 3
# 13 "C:/msys64/mingw64/x86_64-w64-mingw32/include/specstrings.h" 2 3





extern "C" {
# 333 "C:/msys64/mingw64/x86_64-w64-mingw32/include/specstrings.h" 3
}


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/driverspecs.h" 1 3
# 337 "C:/msys64/mingw64/x86_64-w64-mingw32/include/specstrings.h" 2 3
# 7 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 2 3
# 19 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 3
extern "C" {




  typedef unsigned long ULONG;
  typedef ULONG *PULONG;
  typedef unsigned short USHORT;
  typedef USHORT *PUSHORT;
  typedef unsigned char UCHAR;
  typedef UCHAR *PUCHAR;
  typedef char *PSZ;
# 127 "C:/msys64/mingw64/x86_64-w64-mingw32/include/minwindef.h" 3
typedef int WINBOOL;
       


  typedef int BOOL;


typedef WINBOOL *PBOOL;
typedef WINBOOL *LPBOOL;
       


  typedef unsigned char BYTE;
  typedef unsigned short WORD;
  typedef unsigned long DWORD;
  typedef float FLOAT;
  typedef FLOAT *PFLOAT;
  typedef BYTE *PBYTE;
  typedef BYTE *LPBYTE;
  typedef int *PINT;
  typedef int *LPINT;
  typedef WORD *PWORD;
  typedef WORD *LPWORD;
  typedef long *LPLONG;
  typedef DWORD *PDWORD;
  typedef DWORD *LPDWORD;
  typedef void *LPVOID;


  typedef const void *LPCVOID;

  typedef int INT;
  typedef unsigned int UINT;
  typedef unsigned int *PUINT;


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_unicode.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3


extern "C" {


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 16 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 1 3
# 12 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
extern "C" {
# 43 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  extern unsigned short ** __imp__pctype;
# 59 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  extern unsigned short ** __imp__wctype;
# 71 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  extern unsigned short ** __imp__pwctype;
# 80 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  extern const unsigned char __newclmap[];
  extern const unsigned char __newcumap[];
  extern pthreadlocinfo __ptlocinfo;
  extern pthreadmbcinfo __ptmbcinfo;
  extern int __globallocalestatus;
  extern int __locale_changed;
  extern struct threadlocaleinfostruct __initiallocinfo;
  extern _locale_tstruct __initiallocalestructinfo;
  pthreadlocinfo __attribute__((__cdecl__)) __updatetlocinfo(void);
  pthreadmbcinfo __attribute__((__cdecl__)) __updatetmbcinfo(void);
# 108 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isctype(int _C,int _Type);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isctype_l(int _C,int _Type,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isalpha(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isalpha_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isupper(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isupper_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) islower(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _islower_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isdigit(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isdigit_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isxdigit(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isxdigit_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isspace(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isspace_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) ispunct(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _ispunct_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isalnum(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isalnum_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isprint(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isprint_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) isgraph(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isgraph_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) iscntrl(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iscntrl_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) toupper(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) tolower(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _tolower(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _tolower_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _toupper(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _toupper_l(int _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __isascii(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __toascii(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __iscsymf(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __iscsym(int _C);


int __attribute__((__cdecl__)) isblank(int _C);






  int __attribute__((__cdecl__)) iswalpha(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswalpha_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswupper(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswupper_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswlower(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswlower_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswdigit(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswdigit_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswxdigit(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswxdigit_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswspace(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswspace_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswpunct(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswpunct_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswalnum(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswalnum_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswprint(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswprint_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswgraph(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswgraph_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswcntrl(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswcntrl_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswascii(wint_t _C);
  int __attribute__((__cdecl__)) isleadbyte(int _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _isleadbyte_l(int _C,_locale_t _Locale);
  wint_t __attribute__((__cdecl__)) towupper(wint_t _C);
  __attribute__ ((__dllimport__)) wint_t __attribute__((__cdecl__)) _towupper_l(wint_t _C,_locale_t _Locale);
  wint_t __attribute__((__cdecl__)) towlower(wint_t _C);
  __attribute__ ((__dllimport__)) wint_t __attribute__((__cdecl__)) _towlower_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) iswctype(wint_t _C,wctype_t _Type);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswctype_l(wint_t _C,wctype_t _Type,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __iswcsymf(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswcsymf_l(wint_t _C,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) __iswcsym(wint_t _C);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _iswcsym_l(wint_t _C,_locale_t _Locale);
  int __attribute__((__cdecl__)) is_wctype(wint_t _C,wctype_t _Type);


int __attribute__((__cdecl__)) iswblank(wint_t _C);
# 203 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
  extern int * __imp___mb_cur_max;




__attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) ___mb_cur_max_func(void);
# 286 "C:/msys64/mingw64/x86_64-w64-mingw32/include/ctype.h" 3
}
# 17 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/apiset.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/apiset.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/apiset.h" 2 3


extern "C" {
# 105 "C:/msys64/mingw64/x86_64-w64-mingw32/include/apiset.h" 3
}
# 20 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3







# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 1 3
# 65 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
       
# 561 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
extern "C" {
# 595 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
void __faststorefence(void);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __faststorefence(void) {


    __builtin_ia32_sfence();
}





__extension__ void __stosq(unsigned long long *, unsigned long long, size_t);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __stosq(unsigned long long *Dest, unsigned long long Data, size_t Count) { __asm__ __volatile__ ("rep stos{" "q|q" "}" : "+D" (Dest), "+c" (Count) : [Data] "a" (Data) : "memory"); }





__extension__ unsigned char _interlockedbittestandset64(long long volatile *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandset64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock bts{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ unsigned char _interlockedbittestandreset64(long long volatile *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandreset64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock btr{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ unsigned char _interlockedbittestandcomplement64(long long volatile *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandcomplement64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock btc{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ unsigned char InterlockedBitTestAndSet64(volatile long long *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndSet64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock bts{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ unsigned char InterlockedBitTestAndReset64(volatile long long *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndReset64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock btr{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ unsigned char InterlockedBitTestAndComplement64(volatile long long *a, long long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndComplement64(long long volatile *Base, long long Offset) { unsigned char old; __asm__ __volatile__ ("lock btc{q %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "J" "r" (Offset) : "memory" ); return old; }





__extension__ long long _InterlockedAnd64(long long volatile *, long long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedAnd64(volatile long long *Destination, long long Value) { return __sync_fetch_and_and(Destination, Value); }





__extension__ long long _InterlockedOr64(long long volatile *, long long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedOr64(volatile long long *Destination, long long Value) { return __sync_fetch_and_or(Destination, Value); }





__extension__ long long _InterlockedXor64(long long volatile *, long long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedXor64(volatile long long *Destination, long long Value) { return __sync_fetch_and_xor(Destination, Value); }





__extension__ long long _InterlockedIncrement64(long long volatile *Addend);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedIncrement64(long long volatile *Addend) {
    return __sync_add_and_fetch(Addend, 1);
}





__extension__ long long _InterlockedDecrement64(long long volatile *Addend);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedDecrement64(long long volatile *Addend) {
    return __sync_sub_and_fetch(Addend, 1);
}





__extension__ long long _InterlockedExchange64(long long volatile *Target, long long Value);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedExchange64(long long volatile *Target, long long Value) {
    return __sync_lock_test_and_set(Target, Value);
}





__extension__ long long _InterlockedExchangeAdd64(long long volatile *Addend, long long Value);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedExchangeAdd64(long long volatile *Addend, long long Value) {
    return __sync_fetch_and_add(Addend, Value);
}





unsigned char __readgsbyte(unsigned long Offset);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char __readgsbyte(unsigned long Offset) { unsigned char ret; __asm__ ("mov{" "b" " %%" "gs" ":%[offset], %[ret] | %[ret], %%" "gs" ":%[offset]}" : [ret] "=r" (ret) : [offset] "m" ((*(unsigned char *) (size_t) Offset))); return ret; }





unsigned short __readgsword(unsigned long Offset);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned short __readgsword(unsigned long Offset) { unsigned short ret; __asm__ ("mov{" "w" " %%" "gs" ":%[offset], %[ret] | %[ret], %%" "gs" ":%[offset]}" : [ret] "=r" (ret) : [offset] "m" ((*(unsigned short *) (size_t) Offset))); return ret; }





unsigned long __readgsdword(unsigned long Offset);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned long __readgsdword(unsigned long Offset) { unsigned long ret; __asm__ ("mov{" "l" " %%" "gs" ":%[offset], %[ret] | %[ret], %%" "gs" ":%[offset]}" : [ret] "=r" (ret) : [offset] "m" ((*(unsigned long *) (size_t) Offset))); return ret; }





__extension__ unsigned long long __readgsqword(unsigned long Offset);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned long long __readgsqword(unsigned long Offset) { unsigned long long ret; __asm__ ("mov{" "q" " %%" "gs" ":%[offset], %[ret] | %[ret], %%" "gs" ":%[offset]}" : [ret] "=r" (ret) : [offset] "m" ((*(unsigned long long *) (size_t) Offset))); return ret; }





void __writegsbyte(unsigned long Offset,unsigned char Data);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __writegsbyte(unsigned long Offset, unsigned char Data) { __asm__ ("mov{" "b" " %[Data], %%" "gs" ":%[offset] | %%" "gs" ":%[offset], %[Data]}" : [offset] "=m" ((*(unsigned char *) (size_t) Offset)) : [Data] "ri" (Data)); }





void __writegsword(unsigned long Offset,unsigned short Data);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __writegsword(unsigned long Offset, unsigned short Data) { __asm__ ("mov{" "w" " %[Data], %%" "gs" ":%[offset] | %%" "gs" ":%[offset], %[Data]}" : [offset] "=m" ((*(unsigned short *) (size_t) Offset)) : [Data] "ri" (Data)); }





void __writegsdword(unsigned long Offset,unsigned long Data);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __writegsdword(unsigned long Offset, unsigned long Data) { __asm__ ("mov{" "l" " %[Data], %%" "gs" ":%[offset] | %%" "gs" ":%[offset], %[Data]}" : [offset] "=m" ((*(unsigned long *) (size_t) Offset)) : [Data] "ri" (Data)); }





__extension__ void __writegsqword(unsigned long Offset,unsigned long long Data);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __writegsqword(unsigned long Offset, unsigned long long Data) { __asm__ ("mov{" "q" " %[Data], %%" "gs" ":%[offset] | %%" "gs" ":%[offset], %[Data]}" : [offset] "=m" ((*(unsigned long long *) (size_t) Offset)) : [Data] "ri" (Data)); }





__extension__ unsigned char _BitScanForward64(unsigned long *Index, unsigned long long Mask);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _BitScanForward64(unsigned long *Index, unsigned long long Mask) { unsigned long long n; unsigned char old; __asm__ ("bsf{q %[Mask],%[Index] | %[Index],%[Mask]}" : "=@ccnz" (old), [Index] "=r" (n) : [Mask] "r" (Mask)); *Index = n; return old; }





__extension__ unsigned char _BitScanReverse64(unsigned long *Index, unsigned long long Mask);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _BitScanReverse64(unsigned long *Index, unsigned long long Mask) { unsigned long long n; unsigned char old; __asm__ ("bsr{q %[Mask],%[Index] | %[Index],%[Mask]}" : "=@ccnz" (old), [Index] "=r" (n) : [Mask] "r" (Mask)); *Index = n; return old; }





__extension__ unsigned char _bittest64(long long const *a, long long b);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittest64(const long long *Base, long long Offset) { unsigned char old; __asm__ ("bt{" "q" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old) : [Offset] "J" "r" (Offset), [Base] "rm" (*Base) : ); return old; }





__extension__ unsigned char _bittestandset64(long long *a, long long b);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandset64(long long *Base, long long Offset) { unsigned char old; __asm__ ("bts" "{" "q" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "J" "r" (Offset) : ); return old; }





__extension__ unsigned char _bittestandreset64(long long *a, long long b);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandreset64(long long *Base, long long Offset) { unsigned char old; __asm__ ("btr" "{" "q" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "J" "r" (Offset) : ); return old; }





__extension__ unsigned char _bittestandcomplement64(long long *a, long long b);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandcomplement64(long long *Base, long long Offset) { unsigned char old; __asm__ ("btc" "{" "q" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "J" "r" (Offset) : ); return old; }
# 949 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
__extension__ void __movsq(unsigned long long *Dest, unsigned long long const *Source, size_t Count);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __movsq(unsigned long long *Destination, unsigned long long const *Source, size_t Count) { __asm__ __volatile__ ( "rep movs" "q" : "=D" (Destination), "=S" (Source), "=c" (Count) : "0" (Destination), "1" (Source), "2" (Count) : "memory"); }
# 1257 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
long _InterlockedAnd(long volatile *, long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedAnd(volatile long *Destination, long Value) { return __sync_fetch_and_and(Destination, Value); }





long _InterlockedOr(long volatile *, long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedOr(volatile long *Destination, long Value) { return __sync_fetch_and_or(Destination, Value); }





long _InterlockedXor(long volatile *, long);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedXor(volatile long *Destination, long Value) { return __sync_fetch_and_xor(Destination, Value); }





short _InterlockedIncrement16(short volatile *Addend);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
short _InterlockedIncrement16(short volatile *Addend) {
    return __sync_add_and_fetch(Addend, 1);
}





short _InterlockedDecrement16(short volatile *Addend);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
short _InterlockedDecrement16(short volatile *Addend) {
    return __sync_sub_and_fetch(Addend, 1);
}





short _InterlockedCompareExchange16(short volatile *Destination, short ExChange, short Comperand);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
short _InterlockedCompareExchange16(short volatile *Destination, short ExChange, short Comperand) {
    return __sync_val_compare_and_swap(Destination, Comperand, ExChange);
}





long _InterlockedExchangeAdd(long volatile *Addend, long Value);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedExchangeAdd(long volatile *Addend, long Value) {
    return __sync_fetch_and_add(Addend, Value);
}





long _InterlockedCompareExchange(long volatile *Destination, long ExChange, long Comperand);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedCompareExchange(long volatile *Destination, long ExChange, long Comperand) {
    return __sync_val_compare_and_swap(Destination, Comperand, ExChange);
}





long _InterlockedIncrement(long volatile *Addend);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedIncrement(long volatile *Addend) {
   return __sync_add_and_fetch(Addend, 1);
}





long _InterlockedDecrement(long volatile *Addend);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedDecrement(long volatile *Addend) {
   return __sync_sub_and_fetch(Addend, 1);
}





long _InterlockedAdd(long volatile *Addend, long Value);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedAdd(long volatile *Addend, long Value) {
    return __sync_add_and_fetch(Addend, Value);
}





__extension__ long long _InterlockedAdd64(long long volatile *Addend, long long Value);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedAdd64(long long volatile *Addend, long long Value) {
    return __sync_add_and_fetch(Addend, Value);
}





long _InterlockedExchange(long volatile *Target, long Value);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long _InterlockedExchange(long volatile *Target, long Value) {
    return __sync_lock_test_and_set(Target, Value);
}





__extension__ long long _InterlockedCompareExchange64(long long volatile *Destination, long long ExChange, long long Comperand);

__extension__ extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
long long _InterlockedCompareExchange64(long long volatile *Destination, long long ExChange, long long Comperand) {
    return __sync_val_compare_and_swap(Destination, Comperand, ExChange);
}





void *_InterlockedCompareExchangePointer(void * volatile *Destination, void *ExChange, void *Comperand);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void *_InterlockedCompareExchangePointer(void *volatile *Destination, void *ExChange, void *Comperand) {
    return __sync_val_compare_and_swap(Destination, Comperand, ExChange);
}





void *_InterlockedExchangePointer(void *volatile *Target,void *Value);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void *_InterlockedExchangePointer(void *volatile *Target,void *Value) {
    return __sync_lock_test_and_set(Target, Value);
}
# 1431 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
void __int2c(void);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __int2c(void) {
    __asm__ __volatile__("int {$}" "0x2c" :);
}





void __stosb(unsigned char *, unsigned char, size_t);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __stosb(unsigned char *Dest, unsigned char Data, size_t Count) { __asm__ __volatile__ ("rep stos{" "b|b" "}" : "+D" (Dest), "+c" (Count) : [Data] "a" (Data) : "memory"); }





void __stosw(unsigned short *, unsigned short, size_t);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __stosw(unsigned short *Dest, unsigned short Data, size_t Count) { __asm__ __volatile__ ("rep stos{" "w|w" "}" : "+D" (Dest), "+c" (Count) : [Data] "a" (Data) : "memory"); }





void __stosd(unsigned long *, unsigned long, size_t);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __stosd(unsigned long *Dest, unsigned long Data, size_t Count) { __asm__ __volatile__ ("rep stos{" "l|d" "}" : "+D" (Dest), "+c" (Count) : [Data] "a" (Data) : "memory"); }





unsigned char _interlockedbittestandset(long volatile *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandset(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock bts{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char _interlockedbittestandreset(long volatile *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandreset(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock btr{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char _interlockedbittestandcomplement(long volatile *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _interlockedbittestandcomplement(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock btc{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char InterlockedBitTestAndSet(volatile long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndSet(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock bts{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char InterlockedBitTestAndReset(volatile long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndReset(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock btr{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char InterlockedBitTestAndComplement(volatile long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char InterlockedBitTestAndComplement(long volatile *Base, long Offset) { unsigned char old; __asm__ __volatile__ ("lock btc{l %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+m" (*Base) : [Offset] "I" "r" (Offset) : "memory" ); return old; }





unsigned char _BitScanForward(unsigned long *Index, unsigned long Mask);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _BitScanForward(unsigned long *Index, unsigned long Mask) { unsigned long n; unsigned char old; __asm__ ("bsf{l %[Mask],%[Index] | %[Index],%[Mask]}" : "=@ccnz" (old), [Index] "=r" (n) : [Mask] "r" (Mask)); *Index = n; return old; }





unsigned char _BitScanReverse(unsigned long *Index, unsigned long Mask);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _BitScanReverse(unsigned long *Index, unsigned long Mask) { unsigned long n; unsigned char old; __asm__ ("bsr{l %[Mask],%[Index] | %[Index],%[Mask]}" : "=@ccnz" (old), [Index] "=r" (n) : [Mask] "r" (Mask)); *Index = n; return old; }





unsigned char _bittest(long const *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittest(const long *Base, long Offset) { unsigned char old; __asm__ ("bt{" "l" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old) : [Offset] "I" "r" (Offset), [Base] "rm" (*Base) : ); return old; }





unsigned char _bittestandset(long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandset(long *Base, long Offset) { unsigned char old; __asm__ ("bts" "{" "l" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "I" "r" (Offset) : ); return old; }





unsigned char _bittestandreset(long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandreset(long *Base, long Offset) { unsigned char old; __asm__ ("btr" "{" "l" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "I" "r" (Offset) : ); return old; }





unsigned char _bittestandcomplement(long *a, long b);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
unsigned char _bittestandcomplement(long *Base, long Offset) { unsigned char old; __asm__ ("btc" "{" "l" " %[Offset],%[Base] | %[Base],%[Offset]}" : [old] "=@ccc" (old), [Base] "+rm" (*Base) : [Offset] "I" "r" (Offset) : ); return old; }
# 1752 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
void __movsb(unsigned char *Destination, unsigned char const *Source, size_t Count);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __movsb(unsigned char *Destination, unsigned char const *Source, size_t Count) { __asm__ __volatile__ ( "rep movs" "b" : "=D" (Destination), "=S" (Source), "=c" (Count) : "0" (Destination), "1" (Source), "2" (Count) : "memory"); }





void __movsw(unsigned short *Dest, unsigned short const *Source, size_t Count);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __movsw(unsigned short *Destination, unsigned short const *Source, size_t Count) { __asm__ __volatile__ ( "rep movs" "w" : "=D" (Destination), "=S" (Source), "=c" (Count) : "0" (Destination), "1" (Source), "2" (Count) : "memory"); }





void __movsd(unsigned long *Dest, unsigned long const *Source, size_t Count);

extern __inline__ __attribute__((__always_inline__,__gnu_inline__))
void __movsd(unsigned long *Destination, unsigned long const *Source, size_t Count) { __asm__ __volatile__ ( "rep movs" "d" : "=D" (Destination), "=S" (Source), "=c" (Count) : "0" (Destination), "1" (Source), "2" (Count) : "memory"); }
# 1948 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
}
# 1960 "C:/msys64/mingw64/x86_64-w64-mingw32/include/psdk_inc/intrin-impl.h" 3
       
# 28 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3
# 150 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 1 3




# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 6 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 2 3





__extension__ typedef unsigned long long POINTER_64_INT;
# 26 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 3
extern "C" {


  typedef signed char INT8,*PINT8;
  typedef signed short INT16,*PINT16;
  typedef signed int INT32,*PINT32;
  __extension__ typedef signed long long INT64,*PINT64;
  typedef unsigned char UINT8,*PUINT8;
  typedef unsigned short UINT16,*PUINT16;
  typedef unsigned int UINT32,*PUINT32;
  __extension__ typedef unsigned long long UINT64,*PUINT64;
  typedef signed int LONG32,*PLONG32;
  typedef unsigned int ULONG32,*PULONG32;
  typedef unsigned int DWORD32,*PDWORD32;






  __extension__ typedef long long INT_PTR,*PINT_PTR;
  __extension__ typedef unsigned long long UINT_PTR,*PUINT_PTR;
  __extension__ typedef long long LONG_PTR,*PLONG_PTR;
  __extension__ typedef unsigned long long ULONG_PTR,*PULONG_PTR;
# 62 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 3
  __extension__ typedef long long SHANDLE_PTR;
  __extension__ typedef unsigned long long HANDLE_PTR;
  typedef unsigned int UHALF_PTR,*PUHALF_PTR;
  typedef int HALF_PTR,*PHALF_PTR;


  static __inline unsigned long HandleToULong (const void *h) { return ((unsigned long) (ULONG_PTR) h); }
  static __inline long HandleToLong (const void *h) { return ((long) (LONG_PTR) h); }
  static __inline void *ULongToHandle (const unsigned long h) { return ((void *) (UINT_PTR) h); }
  static __inline void *LongToHandle (const long h) { return ((void *) (INT_PTR) h); }
  static __inline unsigned long PtrToUlong (const void *p) { return ((unsigned long) (ULONG_PTR) p); }
  static __inline unsigned int PtrToUint (const void *p) { return ((unsigned int) (UINT_PTR) p); }
  static __inline unsigned short PtrToUshort (const void *p) { return ((unsigned short) (unsigned long) (ULONG_PTR) p); }
  static __inline long PtrToLong (const void *p) { return ((long) (LONG_PTR) p); }
  static __inline int PtrToInt (const void *p) { return ((int) (INT_PTR) p); }
  static __inline short PtrToShort (const void *p) { return ((short) (long) (LONG_PTR) p); }
  static __inline void *IntToPtr (const int i) { return ((void *) (INT_PTR)i); }
  static __inline void *UIntToPtr (const unsigned int ui) { return ((void *) (UINT_PTR)ui); }
  static __inline void *LongToPtr (const long l) { return ((void *) (LONG_PTR)l); }
  static __inline void *ULongToPtr (const unsigned long ul) { return ((void *) (ULONG_PTR)ul); }






  static __inline void *Ptr32ToPtr (const void * p) { return ((void *) (ULONG_PTR) (unsigned long) (ULONG_PTR) p); }
  static __inline void *Handle32ToHandle (const void * h) { return ((void *) (LONG_PTR) (long) (ULONG_PTR) h); }
  static __inline void * PtrToPtr32 (const void *p) { return ((void *) (ULONG_PTR) (unsigned long) (ULONG_PTR) p); }
# 147 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 3
  __extension__ typedef ULONG_PTR SIZE_T,*PSIZE_T;
  __extension__ typedef LONG_PTR SSIZE_T,*PSSIZE_T;
# 191 "C:/msys64/mingw64/x86_64-w64-mingw32/include/basetsd.h" 3
  __extension__ typedef ULONG_PTR DWORD_PTR,*PDWORD_PTR;
  __extension__ typedef long long LONG64,*PLONG64;
  __extension__ typedef unsigned long long ULONG64,*PULONG64;
  __extension__ typedef unsigned long long DWORD64,*PDWORD64;
  __extension__ typedef ULONG_PTR KAFFINITY;
  __extension__ typedef KAFFINITY *PKAFFINITY;


}
# 151 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3
# 257 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef void *PVOID;
  typedef void *PVOID64;
# 285 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef char CHAR;
  typedef short SHORT;
  typedef long LONG;

  typedef int INT;





  typedef wchar_t WCHAR;

  typedef WCHAR *PWCHAR,*LPWCH,*PWCH;
  typedef const WCHAR *LPCWCH,*PCWCH;
  typedef WCHAR *NWPSTR,*LPWSTR,*PWSTR;
  typedef PWSTR *PZPWSTR;
  typedef const PWSTR *PCZPWSTR;
  typedef WCHAR *LPUWSTR,*PUWSTR;
  typedef const WCHAR *LPCWSTR,*PCWSTR;
  typedef PCWSTR *PZPCWSTR;
  typedef const WCHAR *LPCUWSTR,*PCUWSTR;
  typedef WCHAR *PZZWSTR;
  typedef const WCHAR *PCZZWSTR;
  typedef WCHAR *PUZZWSTR;
  typedef const WCHAR *PCUZZWSTR;
  typedef WCHAR *PNZWCH;
  typedef const WCHAR *PCNZWCH;
  typedef WCHAR *PUNZWCH;
  typedef const WCHAR *PCUNZWCH;


  typedef const WCHAR *LPCWCHAR,*PCWCHAR;
  typedef const WCHAR *LPCUWCHAR,*PCUWCHAR;
  typedef unsigned long UCSCHAR;





  typedef UCSCHAR *PUCSCHAR;
  typedef const UCSCHAR *PCUCSCHAR;
  typedef UCSCHAR *PUCSSTR;
  typedef UCSCHAR *PUUCSSTR;
  typedef const UCSCHAR *PCUCSSTR;
  typedef const UCSCHAR *PCUUCSSTR;
  typedef UCSCHAR *PUUCSCHAR;
  typedef const UCSCHAR *PCUUCSCHAR;


  typedef CHAR *PCHAR,*LPCH,*PCH;
  typedef const CHAR *LPCCH,*PCCH;
  typedef CHAR *NPSTR,*LPSTR,*PSTR;
  typedef PSTR *PZPSTR;
  typedef const PSTR *PCZPSTR;
  typedef const CHAR *LPCSTR,*PCSTR;
  typedef PCSTR *PZPCSTR;
  typedef CHAR *PZZSTR;
  typedef const CHAR *PCZZSTR;
  typedef CHAR *PNZCH;
  typedef const CHAR *PCNZCH;




  typedef WCHAR TCHAR, *PTCHAR;
  typedef WCHAR TBYTE, *PTBYTE;


  typedef LPWSTR LPTCH,PTCH;
  typedef LPWSTR PTSTR,LPTSTR;
  typedef LPCWSTR PCTSTR,LPCTSTR;
  typedef LPUWSTR PUTSTR,LPUTSTR;
  typedef LPCUWSTR PCUTSTR,LPCUTSTR;
  typedef LPWSTR LP;
  typedef PZZWSTR PZZTSTR;
  typedef PCZZWSTR PCZZTSTR;
  typedef PUZZWSTR PUZZTSTR;
  typedef PCUZZWSTR PCUZZTSTR;
  typedef PZPWSTR PZPTSTR;
  typedef PNZWCH PNZTCH;
  typedef PCNZWCH PCNZTCH;
  typedef PUNZWCH PUNZTCH;
  typedef PCUNZWCH PCUNZTCH;
# 392 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef SHORT *PSHORT;
  typedef LONG *PLONG;



typedef struct _GROUP_AFFINITY {
  KAFFINITY Mask;
  WORD Group;
  WORD Reserved[3];
} GROUP_AFFINITY, *PGROUP_AFFINITY;



  typedef void *HANDLE;






  typedef HANDLE *PHANDLE;
  typedef BYTE FCHAR;
  typedef WORD FSHORT;
  typedef DWORD FLONG;



  typedef LONG HRESULT;
# 453 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef char CCHAR;


typedef DWORD LCID;

  typedef PDWORD PLCID;


  typedef WORD LANGID;





typedef enum {
  UNSPECIFIED_COMPARTMENT_ID = 0,
  DEFAULT_COMPARTMENT_ID
} COMPARTMENT_ID,*PCOMPARTMENT_ID;
# 482 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
    typedef struct _FLOAT128 {
      __extension__ long long LowPart;
      __extension__ long long HighPart;
  } FLOAT128;

  typedef FLOAT128 *PFLOAT128;


  __extension__ typedef long long LONGLONG;
  __extension__ typedef unsigned long long ULONGLONG;



  typedef LONGLONG *PLONGLONG;
  typedef ULONGLONG *PULONGLONG;
  typedef LONGLONG USN;







  typedef union _LARGE_INTEGER {
    __extension__ struct {
      DWORD LowPart;
      LONG HighPart;
    } ;
    struct {
      DWORD LowPart;
      LONG HighPart;
    } u;

    LONGLONG QuadPart;
  } LARGE_INTEGER;

  typedef LARGE_INTEGER *PLARGE_INTEGER;




  typedef union _ULARGE_INTEGER {
    __extension__ struct {
      DWORD LowPart;
      DWORD HighPart;
    } ;
    struct {
      DWORD LowPart;
      DWORD HighPart;
    } u;

    ULONGLONG QuadPart;
  } ULARGE_INTEGER;

  typedef ULARGE_INTEGER *PULARGE_INTEGER;

  typedef struct _LUID {
    DWORD LowPart;
    LONG HighPart;
  } LUID,*PLUID;




  typedef ULONGLONG DWORDLONG;
  typedef DWORDLONG *PDWORDLONG;
# 556 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  extern "C" {
# 566 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
    unsigned char __attribute__((__cdecl__)) _rotl8(unsigned char Value,unsigned char Shift);
    unsigned short __attribute__((__cdecl__)) _rotl16(unsigned short Value,unsigned char Shift);
    unsigned char __attribute__((__cdecl__)) _rotr8(unsigned char Value,unsigned char Shift);
    unsigned short __attribute__((__cdecl__)) _rotr16(unsigned short Value,unsigned char Shift);







       
       


    unsigned int __attribute__((__cdecl__)) _rotl(unsigned int Value,int Shift);
    unsigned int __attribute__((__cdecl__)) _rotr(unsigned int Value,int Shift);
       
       
       
       


    __extension__ unsigned long long __attribute__((__cdecl__)) _rotl64(unsigned long long Value,int Shift);
    __extension__ unsigned long long __attribute__((__cdecl__)) _rotr64(unsigned long long Value,int Shift);
       
       


  }
# 605 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef BYTE BOOLEAN;

  typedef BOOLEAN *PBOOLEAN;




  typedef struct _LIST_ENTRY {
    struct _LIST_ENTRY *Flink;
    struct _LIST_ENTRY *Blink;
  } LIST_ENTRY,*PLIST_ENTRY,* PRLIST_ENTRY;

  typedef struct _SINGLE_LIST_ENTRY {
    struct _SINGLE_LIST_ENTRY *Next;
  } SINGLE_LIST_ENTRY,*PSINGLE_LIST_ENTRY;

  typedef struct LIST_ENTRY32 {
    DWORD Flink;
    DWORD Blink;
  } LIST_ENTRY32;
  typedef LIST_ENTRY32 *PLIST_ENTRY32;

  typedef struct LIST_ENTRY64 {
    ULONGLONG Flink;
    ULONGLONG Blink;
  } LIST_ENTRY64;
  typedef LIST_ENTRY64 *PLIST_ENTRY64;



# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 2 3
# 19 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 3
typedef struct _GUID {
  unsigned long Data1;
  unsigned short Data2;
  unsigned short Data3;
  unsigned char Data4[8];
} GUID;




extern "C++" {
__extension__ template<typename T> const GUID &__mingw_uuidof();
}
# 72 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 3
typedef GUID *LPGUID;




typedef const GUID *LPCGUID;





typedef GUID IID;
typedef IID *LPIID;






typedef GUID CLSID;


typedef CLSID *LPCLSID;



typedef GUID FMTID;
typedef FMTID *LPFMTID;
# 149 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/string.h" 1 3
# 21 "C:/msys64/mingw64/x86_64-w64-mingw32/include/string.h" 3
extern "C" {
# 45 "C:/msys64/mingw64/x86_64-w64-mingw32/include/string.h" 3
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _memccpy(void *_Dst,const void *_Src,int _Val,size_t _MaxCount);
  void *__attribute__((__cdecl__)) memchr(const void *_Buf ,int _Val,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _memicmp(const void *_Buf1,const void *_Buf2,size_t _Size);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _memicmp_l(const void *_Buf1,const void *_Buf2,size_t _Size,_locale_t _Locale);
  int __attribute__((__cdecl__)) memcmp(const void *_Buf1,const void *_Buf2,size_t _Size);
  void * __attribute__((__cdecl__)) memcpy(void * __restrict__ _Dst,const void * __restrict__ _Src,size_t _Size) ;
  __attribute__((dllimport)) errno_t __attribute__((__cdecl__)) memcpy_s (void *_dest,size_t _numberOfElements,const void *_src,size_t _count);
  void * __attribute__((__cdecl__)) mempcpy (void *_Dst, const void *_Src, size_t _Size);
  void * __attribute__((__cdecl__)) memset(void *_Dst,int _Val,size_t _Size);

  void * __attribute__((__cdecl__)) memccpy(void *_Dst,const void *_Src,int _Val,size_t _Size) ;
  int __attribute__((__cdecl__)) memicmp(const void *_Buf1,const void *_Buf2,size_t _Size) ;


  char * __attribute__((__cdecl__)) _strset(char *_Str,int _Val) ;
  char * __attribute__((__cdecl__)) _strset_l(char *_Str,int _Val,_locale_t _Locale) ;
  char * __attribute__((__cdecl__)) strcpy(char * __restrict__ _Dest,const char * __restrict__ _Source);
  char * __attribute__((__cdecl__)) strcat(char * __restrict__ _Dest,const char * __restrict__ _Source);
  int __attribute__((__cdecl__)) strcmp(const char *_Str1,const char *_Str2);
  size_t __attribute__((__cdecl__)) strlen(const char *_Str);
  size_t __attribute__((__cdecl__)) strnlen(const char *_Str,size_t _MaxCount);
  void *__attribute__((__cdecl__)) memmove(void *_Dst,const void *_Src,size_t _Size) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strdup(const char *_Src);
  char *__attribute__((__cdecl__)) strchr(const char *_Str,int _Val);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _stricmp(const char *_Str1,const char *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strcmpi(const char *_Str1,const char *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _stricmp_l(const char *_Str1,const char *_Str2,_locale_t _Locale);
  int __attribute__((__cdecl__)) strcoll(const char *_Str1,const char *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strcoll_l(const char *_Str1,const char *_Str2,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _stricoll(const char *_Str1,const char *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _stricoll_l(const char *_Str1,const char *_Str2,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strncoll (const char *_Str1,const char *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strncoll_l(const char *_Str1,const char *_Str2,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strnicoll (const char *_Str1,const char *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strnicoll_l(const char *_Str1,const char *_Str2,size_t _MaxCount,_locale_t _Locale);
  size_t __attribute__((__cdecl__)) strcspn(const char *_Str,const char *_Control);
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strerror(const char *_ErrMsg) ;
  char *__attribute__((__cdecl__)) strerror(int) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strlwr(char *_String) ;
  char *strlwr_l(char *_String,_locale_t _Locale) ;
  char *__attribute__((__cdecl__)) strncat(char * __restrict__ _Dest,const char * __restrict__ _Source,size_t _Count) ;
  int __attribute__((__cdecl__)) strncmp(const char *_Str1,const char *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strnicmp(const char *_Str1,const char *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _strnicmp_l(const char *_Str1,const char *_Str2,size_t _MaxCount,_locale_t _Locale);
  char *strncpy(char * __restrict__ _Dest,const char * __restrict__ _Source,size_t _Count) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strnset(char *_Str,int _Val,size_t _MaxCount) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strnset_l(char *str,int c,size_t count,_locale_t _Locale) ;
  char *__attribute__((__cdecl__)) strpbrk(const char *_Str,const char *_Control);
  char *__attribute__((__cdecl__)) strrchr(const char *_Str,int _Ch);
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strrev(char *_Str);
  size_t __attribute__((__cdecl__)) strspn(const char *_Str,const char *_Control);
  char *__attribute__((__cdecl__)) strstr(const char *_Str,const char *_SubStr);
  char *__attribute__((__cdecl__)) strtok(char * __restrict__ _Str,const char * __restrict__ _Delim) ;
       

  char *strtok_r(char * __restrict__ _Str, const char * __restrict__ _Delim, char ** __restrict__ __last);
       
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strupr(char *_String) ;
  __attribute__ ((__dllimport__)) char *_strupr_l(char *_String,_locale_t _Locale) ;
  size_t __attribute__((__cdecl__)) strxfrm(char * __restrict__ _Dst,const char * __restrict__ _Src,size_t _MaxCount);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _strxfrm_l(char * __restrict__ _Dst,const char * __restrict__ _Src,size_t _MaxCount,_locale_t _Locale);


  char *__attribute__((__cdecl__)) strdup(const char *_Src) ;
  int __attribute__((__cdecl__)) strcmpi(const char *_Str1,const char *_Str2) ;
  int __attribute__((__cdecl__)) stricmp(const char *_Str1,const char *_Str2) ;
  char *__attribute__((__cdecl__)) strlwr(char *_Str) ;
  int __attribute__((__cdecl__)) strnicmp(const char *_Str1,const char *_Str,size_t _MaxCount) ;
  int __attribute__((__cdecl__)) strncasecmp (const char *, const char *, size_t);
  int __attribute__((__cdecl__)) strcasecmp (const char *, const char *);







  char *__attribute__((__cdecl__)) strnset(char *_Str,int _Val,size_t _MaxCount) ;
  char *__attribute__((__cdecl__)) strrev(char *_Str) ;
  char *__attribute__((__cdecl__)) strset(char *_Str,int _Val) ;
  char *__attribute__((__cdecl__)) strupr(char *_Str) ;





  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcsdup(const wchar_t *_Str);
  wchar_t *__attribute__((__cdecl__)) wcscat(wchar_t * __restrict__ _Dest,const wchar_t * __restrict__ _Source) ;
  wchar_t *__attribute__((__cdecl__)) wcschr(const wchar_t *_Str,wchar_t _Ch);
  int __attribute__((__cdecl__)) wcscmp(const wchar_t *_Str1,const wchar_t *_Str2);
  wchar_t *__attribute__((__cdecl__)) wcscpy(wchar_t * __restrict__ _Dest,const wchar_t * __restrict__ _Source) ;
  size_t __attribute__((__cdecl__)) wcscspn(const wchar_t *_Str,const wchar_t *_Control);
  size_t __attribute__((__cdecl__)) wcslen(const wchar_t *_Str);
  size_t __attribute__((__cdecl__)) wcsnlen(const wchar_t *_Src,size_t _MaxCount);
  wchar_t *wcsncat(wchar_t * __restrict__ _Dest,const wchar_t * __restrict__ _Source,size_t _Count) ;
  int __attribute__((__cdecl__)) wcsncmp(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount);
  wchar_t *wcsncpy(wchar_t * __restrict__ _Dest,const wchar_t * __restrict__ _Source,size_t _Count) ;
  wchar_t *__attribute__((__cdecl__)) _wcsncpy_l(wchar_t * __restrict__ _Dest,const wchar_t * __restrict__ _Source,size_t _Count,_locale_t _Locale) ;
  wchar_t *__attribute__((__cdecl__)) wcspbrk(const wchar_t *_Str,const wchar_t *_Control);
  wchar_t *__attribute__((__cdecl__)) wcsrchr(const wchar_t *_Str,wchar_t _Ch);
  size_t __attribute__((__cdecl__)) wcsspn(const wchar_t *_Str,const wchar_t *_Control);
  wchar_t *__attribute__((__cdecl__)) wcsstr(const wchar_t *_Str,const wchar_t *_SubStr);
  wchar_t *__attribute__((__cdecl__)) wcstok(wchar_t * __restrict__ _Str,const wchar_t * __restrict__ _Delim) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcserror(int _ErrNum) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) __wcserror(const wchar_t *_Str) ;
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsicmp(const wchar_t *_Str1,const wchar_t *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsicmp_l(const wchar_t *_Str1,const wchar_t *_Str2,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsnicmp(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsnicmp_l(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcsnset(wchar_t *_Str,wchar_t _Val,size_t _MaxCount) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcsrev(wchar_t *_Str);
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcsset(wchar_t *_Str,wchar_t _Val) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcslwr(wchar_t *_String) ;
  __attribute__ ((__dllimport__)) wchar_t *_wcslwr_l(wchar_t *_String,_locale_t _Locale) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcsupr(wchar_t *_String) ;
  __attribute__ ((__dllimport__)) wchar_t *_wcsupr_l(wchar_t *_String,_locale_t _Locale) ;
  size_t __attribute__((__cdecl__)) wcsxfrm(wchar_t * __restrict__ _Dst,const wchar_t * __restrict__ _Src,size_t _MaxCount);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _wcsxfrm_l(wchar_t * __restrict__ _Dst,const wchar_t * __restrict__ _Src,size_t _MaxCount,_locale_t _Locale);
  int __attribute__((__cdecl__)) wcscoll(const wchar_t *_Str1,const wchar_t *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcscoll_l(const wchar_t *_Str1,const wchar_t *_Str2,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsicoll(const wchar_t *_Str1,const wchar_t *_Str2);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsicoll_l(const wchar_t *_Str1,const wchar_t *_Str2,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsncoll(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsncoll_l(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsnicoll(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wcsnicoll_l(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount,_locale_t _Locale);


  wchar_t *__attribute__((__cdecl__)) wcsdup(const wchar_t *_Str) ;

  int __attribute__((__cdecl__)) wcsicmp(const wchar_t *_Str1,const wchar_t *_Str2) ;
  int __attribute__((__cdecl__)) wcsnicmp(const wchar_t *_Str1,const wchar_t *_Str2,size_t _MaxCount) ;
  wchar_t *__attribute__((__cdecl__)) wcsnset(wchar_t *_Str,wchar_t _Val,size_t _MaxCount) ;
  wchar_t *__attribute__((__cdecl__)) wcsrev(wchar_t *_Str) ;
  wchar_t *__attribute__((__cdecl__)) wcsset(wchar_t *_Str,wchar_t _Val) ;
  wchar_t *__attribute__((__cdecl__)) wcslwr(wchar_t *_Str) ;
  wchar_t *__attribute__((__cdecl__)) wcsupr(wchar_t *_Str) ;
  int __attribute__((__cdecl__)) wcsicoll(const wchar_t *_Str1,const wchar_t *_Str2) ;




}


# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/string_s.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/string_s.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/string.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/string_s.h" 2 3
# 23 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/string_s.h" 3
extern "C" {


  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strset_s(char *_Dst,size_t _DstSize,int _Value);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strerror_s(char *_Buf,size_t _SizeInBytes,const char *_ErrMsg);
  __attribute__((dllimport)) errno_t __attribute__((__cdecl__)) strerror_s(char *_Buf,size_t _SizeInBytes,int _ErrNum);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strlwr_s(char *_Str,size_t _Size);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strlwr_s_l(char *_Str,size_t _Size,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strnset_s(char *_Str,size_t _Size,int _Val,size_t _MaxCount);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strupr_s(char *_Str,size_t _Size);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strupr_s_l(char *_Str,size_t _Size,_locale_t _Locale);

  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) strncat_s(char *_Dst,size_t _DstSizeInChars,const char *_Src,size_t _MaxCount);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strncat_s_l(char *_Dst,size_t _DstSizeInChars,const char *_Src,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) strcpy_s(char *_Dst, rsize_t _SizeInBytes, const char *_Src);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) strcpy_s(char (&_Dest)[__size], const char * _Source) { return strcpy_s(_Dest,__size,_Source); } }
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) strncpy_s(char *_Dst, size_t _DstSizeInChars, const char *_Src, size_t _MaxCount);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) strncpy_s(char (&_Dest)[__size], const char * _Source, size_t _MaxCount) { return strncpy_s(_Dest,__size,_Source,_MaxCount); } }
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _strncpy_s_l(char *_Dst, size_t _DstSizeInChars, const char *_Src, size_t _MaxCount, _locale_t _Locale);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) _strncpy_s_l(char (&_Dest)[__size], const char * _Source, size_t _MaxCount, _locale_t _Locale) { return _strncpy_s_l(_Dest,__size,_Source,_MaxCount,_Locale); } };
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) strtok_s(char *_Str,const char *_Delim,char **_Context);
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _strtok_s_l(char *_Str,const char *_Delim,char **_Context,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) strcat_s(char *_Dst, rsize_t _SizeInBytes, const char * _Src);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) strcat_s(char (&_Dest)[__size], const char * _Source) { return strcat_s(_Dest,__size,_Source); } }

  __attribute__((dllimport)) errno_t __attribute__((__cdecl__)) memmove_s(void *_dest,size_t _numberOfElements,const void *_src,size_t _count);


  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) wcstok_s(wchar_t *_Str,const wchar_t *_Delim,wchar_t **_Context);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcserror_s(wchar_t *_Buf,size_t _SizeInWords,int _ErrNum);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) __wcserror_s(wchar_t *_Buffer,size_t _SizeInWords,const wchar_t *_ErrMsg);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsnset_s(wchar_t *_Dst,size_t _DstSizeInWords,wchar_t _Val,size_t _MaxCount);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsset_s(wchar_t *_Str,size_t _SizeInWords,wchar_t _Val);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcslwr_s(wchar_t *_Str,size_t _SizeInWords);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcslwr_s_l(wchar_t *_Str,size_t _SizeInWords,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsupr_s(wchar_t *_Str,size_t _Size);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsupr_s_l(wchar_t *_Str,size_t _Size,_locale_t _Locale);

  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) wcscpy_s(wchar_t *_Dst, rsize_t _SizeInWords, const wchar_t *_Src);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) wcscpy_s(wchar_t (&_Dest)[__size], const wchar_t * _Source) { return wcscpy_s(_Dest,__size,_Source); } }
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) wcscat_s(wchar_t * _Dst, rsize_t _SizeInWords, const wchar_t *_Src);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) wcscat_s(wchar_t (&_Dest)[__size], const wchar_t * _Source) { return wcscat_s(_Dest,__size,_Source); } }

  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) wcsncat_s(wchar_t *_Dst,size_t _DstSizeInChars,const wchar_t *_Src,size_t _MaxCount);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsncat_s_l(wchar_t *_Dst,size_t _DstSizeInChars,const wchar_t *_Src,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) wcsncpy_s(wchar_t *_Dst, size_t _DstSizeInChars, const wchar_t *_Src, size_t _MaxCount);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) wcsncpy_s(wchar_t (&_Dest)[__size], const wchar_t * _Source, size_t _MaxCount) { return wcsncpy_s(_Dest,__size,_Source,_MaxCount); } };
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsncpy_s_l(wchar_t *_Dst, size_t _DstSizeInChars, const wchar_t *_Src, size_t _MaxCount, _locale_t _Locale);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) _wcsncpy_s_l(wchar_t (&_Dest)[__size], const wchar_t * _Source, size_t _MaxCount, _locale_t _Locale) { return _wcsncpy_s_l(_Dest,__size,_Source,_MaxCount,_Locale); } };
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wcstok_s_l(wchar_t *_Str,const wchar_t *_Delim,wchar_t **_Context,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsset_s_l(wchar_t *_Str,size_t _SizeInChars,unsigned int _Val,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcsnset_s_l(wchar_t *_Str,size_t _SizeInChars,unsigned int _Val, size_t _Count,_locale_t _Locale);

  inline __attribute__((__always_inline__)) size_t __attribute__((__cdecl__)) wcsnlen_s(const wchar_t * _src, size_t _count) {
    return _src ? wcsnlen(_src, _count) : 0;
  }



}
# 191 "C:/msys64/mingw64/x86_64-w64-mingw32/include/string.h" 2 3
# 150 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 2 3


__inline int InlineIsEqualGUID (const GUID & rguid1, const GUID & rguid2) {
  return ((&rguid1.Data1)[0] == (&rguid2.Data1)[0] && (&rguid1.Data1)[1] == (&rguid2.Data1)[1] && (&rguid1.Data1)[2] == (&rguid2.Data1)[2] && (&rguid1.Data1)[3] == (&rguid2.Data1)[3]);
}

__inline int IsEqualGUID (const GUID & rguid1, const GUID & rguid2) {
  return !memcmp (&rguid1,&rguid2, sizeof (GUID));
}
# 175 "C:/msys64/mingw64/x86_64-w64-mingw32/include/guiddef.h" 3
__inline bool operator== (const GUID & guidOne, const GUID & guidOther) { return !!IsEqualGUID (guidOne, guidOther); }
__inline bool operator!= (const GUID & guidOne, const GUID & guidOther) { return ! (guidOne == guidOther); }
# 636 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 2 3



  typedef struct _OBJECTID {
    GUID Lineage;
    DWORD Uniquifier;
  } OBJECTID;
# 706 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
    typedef int EXCEPTION_ROUTINE (struct _EXCEPTION_RECORD *ExceptionRecord, PVOID EstablisherFrame, struct _CONTEXT *ContextRecord, PVOID DispatcherContext);


    typedef EXCEPTION_ROUTINE *PEXCEPTION_ROUTINE;
# 1383 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
  typedef ULONG_PTR KSPIN_LOCK;
  typedef KSPIN_LOCK *PKSPIN_LOCK;

    typedef struct __attribute__ ((__aligned__ (16))) _M128A {
      ULONGLONG Low;
      LONGLONG High;
    } M128A,*PM128A;

    typedef struct __attribute__ ((__aligned__ (16))) _XSAVE_FORMAT {
      WORD ControlWord;
      WORD StatusWord;
      BYTE TagWord;
      BYTE Reserved1;
      WORD ErrorOpcode;
      DWORD ErrorOffset;
      WORD ErrorSelector;
      WORD Reserved2;
      DWORD DataOffset;
      WORD DataSelector;
      WORD Reserved3;
      DWORD MxCsr;
      DWORD MxCsr_Mask;
      M128A FloatRegisters[8];

      M128A XmmRegisters[16];
      BYTE Reserved4[96];





    } XSAVE_FORMAT,*PXSAVE_FORMAT;

    typedef struct __attribute__ ((__aligned__ (8))) _XSAVE_AREA_HEADER {
      DWORD64 Mask;
      DWORD64 Reserved[7];
    } XSAVE_AREA_HEADER,*PXSAVE_AREA_HEADER;

    typedef struct __attribute__ ((__aligned__ (16))) _XSAVE_AREA {
      XSAVE_FORMAT LegacyState;
      XSAVE_AREA_HEADER Header;
    } XSAVE_AREA,*PXSAVE_AREA;

    typedef struct _XSTATE_CONTEXT {
      DWORD64 Mask;
      DWORD Length;
      DWORD Reserved1;
      PXSAVE_AREA Area;



      PVOID Buffer;



    } XSTATE_CONTEXT,*PXSTATE_CONTEXT;

    typedef struct _SCOPE_TABLE_AMD64 {
      DWORD Count;
      struct {
 DWORD BeginAddress;
 DWORD EndAddress;
 DWORD HandlerAddress;
 DWORD JumpTarget;
      } ScopeRecord[1];
    } SCOPE_TABLE_AMD64,*PSCOPE_TABLE_AMD64;






  extern "C" {
# 1552 "C:/msys64/mingw64/x86_64-w64-mingw32/include/winnt.h" 3
extern "C" {

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 1 3 4
# 27 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 3 4
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ia32intrin.h" 1 3 4
# 29 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ia32intrin.h" 3 4
extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bsfd (int __X)
{
  return __builtin_ctz (__X);
}


extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bsrd (int __X)
{
  return __builtin_ia32_bsrsi (__X);
}


extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bswapd (int __X)
{
  return __builtin_bswap32 (__X);
}




#pragma GCC push_options
#pragma GCC target("sse4.2")




extern __inline unsigned int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__crc32b (unsigned int __C, unsigned char __V)
{
  return __builtin_ia32_crc32qi (__C, __V);
}

extern __inline unsigned int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__crc32w (unsigned int __C, unsigned short __V)
{
  return __builtin_ia32_crc32hi (__C, __V);
}

extern __inline unsigned int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__crc32d (unsigned int __C, unsigned int __V)
{
  return __builtin_ia32_crc32si (__C, __V);
}



#pragma GCC pop_options





extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__popcntd (unsigned int __X)
{
  return __builtin_popcount (__X);
}




extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rdpmc (int __S)
{
  return __builtin_ia32_rdpmc (__S);
}




extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rdtsc (void)
{
  return __builtin_ia32_rdtsc ();
}




extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rdtscp (unsigned int *__A)
{
  return __builtin_ia32_rdtscp (__A);
}




extern __inline unsigned char
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rolb (unsigned char __X, int __C)
{
  return __builtin_ia32_rolqi (__X, __C);
}


extern __inline unsigned short
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rolw (unsigned short __X, int __C)
{
  return __builtin_ia32_rolhi (__X, __C);
}


extern __inline unsigned int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rold (unsigned int __X, int __C)
{
  return (__X << __C) | (__X >> (32 - __C));
}


extern __inline unsigned char
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rorb (unsigned char __X, int __C)
{
  return __builtin_ia32_rorqi (__X, __C);
}


extern __inline unsigned short
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rorw (unsigned short __X, int __C)
{
  return __builtin_ia32_rorhi (__X, __C);
}


extern __inline unsigned int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rord (unsigned int __X, int __C)
{
  return (__X >> __C) | (__X << (32 - __C));
}


extern __inline void
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__pause (void)
{
  __builtin_ia32_pause ();
}



extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bsfq (long long __X)
{
  return __builtin_ctzll (__X);
}


extern __inline int
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bsrq (long long __X)
{
  return __builtin_ia32_bsrdi (__X);
}


extern __inline long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__bswapq (long long __X)
{
  return __builtin_bswap64 (__X);
}


#pragma GCC push_options
#pragma GCC target("sse4.2")




extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__crc32q (unsigned long long __C, unsigned long long __V)
{
  return __builtin_ia32_crc32di (__C, __V);
}



#pragma GCC pop_options



extern __inline long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__popcntq (unsigned long long __X)
{
  return __builtin_popcountll (__X);
}


extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rolq (unsigned long long __X, int __C)
{
  return (__X << __C) | (__X >> (64 - __C));
}


extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__rorq (unsigned long long __X, int __C)
{
  return (__X >> __C) | (__X << (64 - __C));
}


extern __inline unsigned long long
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__readeflags (void)
{
  return __builtin_ia32_readeflags_u64 ();
}


extern __inline void
__attribute__((__gnu_inline__, __always_inline__, __artificial__))
__writeeflags (unsigned long long __X)
{
  __builtin_ia32_writeeflags_u64 (__X);
}
# 28 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4



# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mmintrin.h" 1 3 4
# 42 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mmintrin.h" 3 4
typedef int __m64 __attribute__ ((__vector_size__ (8), __may_alias__));


typedef int __m64_u __attribute__ ((__vector_size__ (8), __may_alias__, __aligned__ (1)));


typedef int __v2si __attribute__ ((__vector_size__ (8)));
typedef short __v4hi __attribute__ ((__vector_size__ (8)));
typedef char __v8qi __attribute__ ((__vector_size__ (8)));
typedef long long __v1di __attribute__ ((__vector_size__ (8)));
typedef float __v2sf __attribute__ ((__vector_size__ (8)));


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_empty (void)
{
  __builtin_ia32_emms ();
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_empty (void)
{
  _mm_empty ();
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi32_si64 (int __i)
{
  return (__m64) __builtin_ia32_vec_init_v2si (__i, 0);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_from_int (int __i)
{
  return _mm_cvtsi32_si64 (__i);
}





extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_from_int64 (long long __i)
{
  return (__m64) __i;
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_m64 (long long __i)
{
  return (__m64) __i;
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64x_si64 (long long __i)
{
  return (__m64) __i;
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi64x (long long __i)
{
  return (__m64) __i;
}



extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_si32 (__m64 __i)
{
  return __builtin_ia32_vec_ext_v2si ((__v2si)__i, 0);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_to_int (__m64 __i)
{
  return _mm_cvtsi64_si32 (__i);
}





extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_to_int64 (__m64 __i)
{
  return (long long)__i;
}

extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtm64_si64 (__m64 __i)
{
  return (long long)__i;
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_si64x (__m64 __i)
{
  return (long long)__i;
}





extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packs_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_packsswb ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_packsswb (__m64 __m1, __m64 __m2)
{
  return _mm_packs_pi16 (__m1, __m2);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packs_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_packssdw ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_packssdw (__m64 __m1, __m64 __m2)
{
  return _mm_packs_pi32 (__m1, __m2);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packs_pu16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_packuswb ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_packuswb (__m64 __m1, __m64 __m2)
{
  return _mm_packs_pu16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpckhbw ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpckhbw (__m64 __m1, __m64 __m2)
{
  return _mm_unpackhi_pi8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpckhwd ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpckhwd (__m64 __m1, __m64 __m2)
{
  return _mm_unpackhi_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpckhdq ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpckhdq (__m64 __m1, __m64 __m2)
{
  return _mm_unpackhi_pi32 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpcklbw ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpcklbw (__m64 __m1, __m64 __m2)
{
  return _mm_unpacklo_pi8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpcklwd ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpcklwd (__m64 __m1, __m64 __m2)
{
  return _mm_unpacklo_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_punpckldq ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_punpckldq (__m64 __m1, __m64 __m2)
{
  return _mm_unpacklo_pi32 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddb (__m64 __m1, __m64 __m2)
{
  return _mm_add_pi8 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddw (__m64 __m1, __m64 __m2)
{
  return _mm_add_pi16 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddd ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddd (__m64 __m1, __m64 __m2)
{
  return _mm_add_pi32 (__m1, __m2);
}
# 322 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mmintrin.h" 3 4
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_si64 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddq ((__v1di)__m1, (__v1di)__m2);
}







extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddsb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddsb (__m64 __m1, __m64 __m2)
{
  return _mm_adds_pi8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddsw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddsw (__m64 __m1, __m64 __m2)
{
  return _mm_adds_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_pu8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddusb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddusb (__m64 __m1, __m64 __m2)
{
  return _mm_adds_pu8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_pu16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_paddusw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_paddusw (__m64 __m1, __m64 __m2)
{
  return _mm_adds_pu16 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubb (__m64 __m1, __m64 __m2)
{
  return _mm_sub_pi8 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubw (__m64 __m1, __m64 __m2)
{
  return _mm_sub_pi16 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubd ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubd (__m64 __m1, __m64 __m2)
{
  return _mm_sub_pi32 (__m1, __m2);
}
# 434 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mmintrin.h" 3 4
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_si64 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubq ((__v1di)__m1, (__v1di)__m2);
}







extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubsb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubsb (__m64 __m1, __m64 __m2)
{
  return _mm_subs_pi8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubsw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubsw (__m64 __m1, __m64 __m2)
{
  return _mm_subs_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_pu8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubusb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubusb (__m64 __m1, __m64 __m2)
{
  return _mm_subs_pu8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_pu16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_psubusw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psubusw (__m64 __m1, __m64 __m2)
{
  return _mm_subs_pu16 (__m1, __m2);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_madd_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pmaddwd ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmaddwd (__m64 __m1, __m64 __m2)
{
  return _mm_madd_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhi_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pmulhw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmulhw (__m64 __m1, __m64 __m2)
{
  return _mm_mulhi_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mullo_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pmullw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmullw (__m64 __m1, __m64 __m2)
{
  return _mm_mullo_pi16 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_pi16 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (__v4hi)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psllw (__m64 __m, __m64 __count)
{
  return _mm_sll_pi16 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_pi16 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psllwi ((__v4hi)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psllwi (__m64 __m, int __count)
{
  return _mm_slli_pi16 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_pi32 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_pslld ((__v2si)__m, (__v2si)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pslld (__m64 __m, __m64 __count)
{
  return _mm_sll_pi32 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_pi32 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_pslldi ((__v2si)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pslldi (__m64 __m, int __count)
{
  return _mm_slli_pi32 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_si64 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psllq ((__v1di)__m, (__v1di)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psllq (__m64 __m, __m64 __count)
{
  return _mm_sll_si64 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_si64 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psllqi ((__v1di)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psllqi (__m64 __m, int __count)
{
  return _mm_slli_si64 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sra_pi16 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (__v4hi)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psraw (__m64 __m, __m64 __count)
{
  return _mm_sra_pi16 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srai_pi16 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psrawi ((__v4hi)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrawi (__m64 __m, int __count)
{
  return _mm_srai_pi16 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sra_pi32 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psrad ((__v2si)__m, (__v2si)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrad (__m64 __m, __m64 __count)
{
  return _mm_sra_pi32 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srai_pi32 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psradi ((__v2si)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psradi (__m64 __m, int __count)
{
  return _mm_srai_pi32 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_pi16 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (__v4hi)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrlw (__m64 __m, __m64 __count)
{
  return _mm_srl_pi16 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_pi16 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psrlwi ((__v4hi)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrlwi (__m64 __m, int __count)
{
  return _mm_srli_pi16 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_pi32 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psrld ((__v2si)__m, (__v2si)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrld (__m64 __m, __m64 __count)
{
  return _mm_srl_pi32 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_pi32 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psrldi ((__v2si)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrldi (__m64 __m, int __count)
{
  return _mm_srli_pi32 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_si64 (__m64 __m, __m64 __count)
{
  return (__m64) __builtin_ia32_psrlq ((__v1di)__m, (__v1di)__count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrlq (__m64 __m, __m64 __count)
{
  return _mm_srl_si64 (__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_si64 (__m64 __m, int __count)
{
  return (__m64) __builtin_ia32_psrlqi ((__v1di)__m, __count);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psrlqi (__m64 __m, int __count)
{
  return _mm_srli_si64 (__m, __count);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_and_si64 (__m64 __m1, __m64 __m2)
{
  return __builtin_ia32_pand (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pand (__m64 __m1, __m64 __m2)
{
  return _mm_and_si64 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_andnot_si64 (__m64 __m1, __m64 __m2)
{
  return __builtin_ia32_pandn (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pandn (__m64 __m1, __m64 __m2)
{
  return _mm_andnot_si64 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_or_si64 (__m64 __m1, __m64 __m2)
{
  return __builtin_ia32_por (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_por (__m64 __m1, __m64 __m2)
{
  return _mm_or_si64 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_xor_si64 (__m64 __m1, __m64 __m2)
{
  return __builtin_ia32_pxor (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pxor (__m64 __m1, __m64 __m2)
{
  return _mm_xor_si64 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpeqb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpeqb (__m64 __m1, __m64 __m2)
{
  return _mm_cmpeq_pi8 (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_pi8 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpgtb ((__v8qi)__m1, (__v8qi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpgtb (__m64 __m1, __m64 __m2)
{
  return _mm_cmpgt_pi8 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpeqw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpeqw (__m64 __m1, __m64 __m2)
{
  return _mm_cmpeq_pi16 (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_pi16 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpgtw ((__v4hi)__m1, (__v4hi)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpgtw (__m64 __m1, __m64 __m2)
{
  return _mm_cmpgt_pi16 (__m1, __m2);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpeqd ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpeqd (__m64 __m1, __m64 __m2)
{
  return _mm_cmpeq_pi32 (__m1, __m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_pi32 (__m64 __m1, __m64 __m2)
{
  return (__m64) __builtin_ia32_pcmpgtd ((__v2si)__m1, (__v2si)__m2);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pcmpgtd (__m64 __m1, __m64 __m2)
{
  return _mm_cmpgt_pi32 (__m1, __m2);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setzero_si64 (void)
{
  return (__m64)0LL;
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi32 (int __i1, int __i0)
{
  return (__m64) __builtin_ia32_vec_init_v2si (__i0, __i1);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi16 (short __w3, short __w2, short __w1, short __w0)
{
  return (__m64) __builtin_ia32_vec_init_v4hi (__w0, __w1, __w2, __w3);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pi8 (char __b7, char __b6, char __b5, char __b4,
      char __b3, char __b2, char __b1, char __b0)
{
  return (__m64) __builtin_ia32_vec_init_v8qi (__b0, __b1, __b2, __b3,
            __b4, __b5, __b6, __b7);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_pi32 (int __i0, int __i1)
{
  return _mm_set_pi32 (__i1, __i0);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_pi16 (short __w0, short __w1, short __w2, short __w3)
{
  return _mm_set_pi16 (__w3, __w2, __w1, __w0);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_pi8 (char __b0, char __b1, char __b2, char __b3,
       char __b4, char __b5, char __b6, char __b7)
{
  return _mm_set_pi8 (__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_pi32 (int __i)
{
  return _mm_set_pi32 (__i, __i);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_pi16 (short __w)
{
  return _mm_set_pi16 (__w, __w, __w, __w);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_pi8 (char __b)
{
  return _mm_set_pi8 (__b, __b, __b, __b, __b, __b, __b, __b);
}
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 1 3 4
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mm_malloc.h" 1 3 4
# 27 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mm_malloc.h" 3 4
# 1 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 1 3 4
# 36 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 3 4
# 1 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 1 3 4
# 39 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3 4
       
# 40 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3

# 1 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 1 3
# 229 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 3
namespace std
{
  typedef long long unsigned int size_t;
  typedef long long int ptrdiff_t;


  typedef decltype(nullptr) nullptr_t;

}
# 251 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 3
namespace std
{
  inline namespace __cxx11 __attribute__((__abi_tag__ ("cxx11"))) { }
}
namespace __gnu_cxx
{
  inline namespace __cxx11 __attribute__((__abi_tag__ ("cxx11"))) { }
}
# 533 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 3
# 1 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/os_defines.h" 1 3
# 534 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 2 3


# 1 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/cpu_defines.h" 1 3
# 537 "C:/msys64/mingw64/include/c++/7.3.0/x86_64-w64-mingw32/bits/c++config.h" 2 3
# 42 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 2 3
# 75 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 1 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 1 3 4
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 3 4
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/syslimits.h" 1 3 4






# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 1 3 4
# 194 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 3 4
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/limits.h" 1 3 4
# 195 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 2 3 4
# 8 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/syslimits.h" 2 3 4
# 35 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include-fixed/limits.h" 2 3 4
# 11 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 2 3
# 25 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
#pragma pack(push,_CRT_PACKING)


extern "C" {
# 49 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  typedef int (__attribute__((__cdecl__)) *_onexit_t)(void);
# 59 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  typedef struct _div_t {
    int quot;
    int rem;
  } div_t;

  typedef struct _ldiv_t {
    long quot;
    long rem;
  } ldiv_t;





#pragma pack(4)
  typedef struct {
    unsigned char ld[10];
  } _LDOUBLE;
#pragma pack()



  typedef struct {
    double x;
  } _CRT_DOUBLE;

  typedef struct {
    float f;
  } _CRT_FLOAT;

       


  typedef struct {
    long double x;
  } _LONGDOUBLE;

       

#pragma pack(4)
  typedef struct {
    unsigned char ld12[12];
  } _LDBL12;
#pragma pack()
# 142 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  typedef void (__attribute__((__cdecl__)) *_purecall_handler)(void);

  __attribute__ ((__dllimport__)) _purecall_handler __attribute__((__cdecl__)) _set_purecall_handler(_purecall_handler _Handler);
  __attribute__ ((__dllimport__)) _purecall_handler __attribute__((__cdecl__)) _get_purecall_handler(void);

  typedef void (__attribute__((__cdecl__)) *_invalid_parameter_handler)(const wchar_t *,const wchar_t *,const wchar_t *,unsigned int,uintptr_t);
  __attribute__ ((__dllimport__)) _invalid_parameter_handler __attribute__((__cdecl__)) _set_invalid_parameter_handler(_invalid_parameter_handler _Handler);
  __attribute__ ((__dllimport__)) _invalid_parameter_handler __attribute__((__cdecl__)) _get_invalid_parameter_handler(void);



  __attribute__ ((__dllimport__)) extern int *__attribute__((__cdecl__)) _errno(void);

  errno_t __attribute__((__cdecl__)) _set_errno(int _Value);
  errno_t __attribute__((__cdecl__)) _get_errno(int *_Value);

  __attribute__ ((__dllimport__)) unsigned long *__attribute__((__cdecl__)) __doserrno(void);

  errno_t __attribute__((__cdecl__)) _set_doserrno(unsigned long _Value);
  errno_t __attribute__((__cdecl__)) _get_doserrno(unsigned long *_Value);
# 172 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  extern __attribute__((dllimport)) char *_sys_errlist[1];
  extern __attribute__((dllimport)) int _sys_nerr;





  __attribute__ ((__dllimport__)) char ***__attribute__((__cdecl__)) __p___argv(void);
  __attribute__ ((__dllimport__)) int *__attribute__((__cdecl__)) __p__fmode(void);
# 190 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  errno_t __attribute__((__cdecl__)) _get_pgmptr(char **_Value);
  errno_t __attribute__((__cdecl__)) _get_wpgmptr(wchar_t **_Value);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _set_fmode(int _Mode);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _get_fmode(int *_PMode);
# 281 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  extern int * __imp___argc;



  extern char *** __imp___argv;



  extern wchar_t *** __imp___wargv;





  extern char *** __imp__environ;




  extern wchar_t *** __imp__wenviron;





  extern char ** __imp__pgmptr;




  extern wchar_t ** __imp__wpgmptr;




  extern unsigned int * __imp__osplatform;




  extern unsigned int * __imp__osver;




  extern unsigned int * __imp__winver;




  extern unsigned int * __imp__winmajor;




  extern unsigned int * __imp__winminor;





  errno_t __attribute__((__cdecl__)) _get_osplatform(unsigned int *_Value);
  errno_t __attribute__((__cdecl__)) _get_osver(unsigned int *_Value);
  errno_t __attribute__((__cdecl__)) _get_winver(unsigned int *_Value);
  errno_t __attribute__((__cdecl__)) _get_winmajor(unsigned int *_Value);
  errno_t __attribute__((__cdecl__)) _get_winminor(unsigned int *_Value);




  extern "C++" {
    template <typename _CountofType,size_t _SizeOfArray> char (*__countof_helper( _CountofType (&_Array)[_SizeOfArray]))[_SizeOfArray];

  }





  void __attribute__((__cdecl__)) __attribute__ ((__nothrow__)) exit(int _Code) __attribute__ ((__noreturn__));
  void __attribute__((__cdecl__)) __attribute__ ((__nothrow__)) _exit(int _Code) __attribute__ ((__noreturn__));



  void __attribute__((__cdecl__)) _Exit(int) __attribute__ ((__noreturn__));






       

  void __attribute__((__cdecl__)) __attribute__ ((__noreturn__)) abort(void);
       



  __attribute__ ((__dllimport__)) unsigned int __attribute__((__cdecl__)) _set_abort_behavior(unsigned int _Flags,unsigned int _Mask);



  int __attribute__((__cdecl__)) abs(int _X);
  long __attribute__((__cdecl__)) labs(long _X);


  __extension__ long long __attribute__((__cdecl__)) _abs64(long long);

  extern __inline__ __attribute__((__always_inline__,__gnu_inline__)) long long __attribute__((__cdecl__)) _abs64(long long x) {
    return __builtin_llabs(x);
  }


  int __attribute__((__cdecl__)) atexit(void (__attribute__((__cdecl__)) *)(void));


  double __attribute__((__cdecl__)) atof(const char *_String);
  double __attribute__((__cdecl__)) _atof_l(const char *_String,_locale_t _Locale);

  int __attribute__((__cdecl__)) atoi(const char *_Str);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atoi_l(const char *_Str,_locale_t _Locale);
  long __attribute__((__cdecl__)) atol(const char *_Str);
  __attribute__ ((__dllimport__)) long __attribute__((__cdecl__)) _atol_l(const char *_Str,_locale_t _Locale);


  void *__attribute__((__cdecl__)) bsearch(const void *_Key,const void *_Base,size_t _NumOfElements,size_t _SizeOfElements,int (__attribute__((__cdecl__)) *_PtFuncCompare)(const void *,const void *));
  void __attribute__((__cdecl__)) qsort(void *_Base,size_t _NumOfElements,size_t _SizeOfElements,int (__attribute__((__cdecl__)) *_PtFuncCompare)(const void *,const void *));

  unsigned short __attribute__((__cdecl__)) _byteswap_ushort(unsigned short _Short);
  unsigned long __attribute__((__cdecl__)) _byteswap_ulong (unsigned long _Long);
  __extension__ unsigned long long __attribute__((__cdecl__)) _byteswap_uint64(unsigned long long _Int64);
  div_t __attribute__((__cdecl__)) div(int _Numerator,int _Denominator);
  char *__attribute__((__cdecl__)) getenv(const char *_VarName) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _itoa(int _Value,char *_Dest,int _Radix);
  __extension__ __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _i64toa(long long _Val,char *_DstBuf,int _Radix) ;
  __extension__ __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _ui64toa(unsigned long long _Val,char *_DstBuf,int _Radix) ;
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _atoi64(const char *_String);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _atoi64_l(const char *_String,_locale_t _Locale);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _strtoi64(const char *_String,char **_EndPtr,int _Radix);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _strtoi64_l(const char *_String,char **_EndPtr,int _Radix,_locale_t _Locale);
  __extension__ __attribute__ ((__dllimport__)) unsigned long long __attribute__((__cdecl__)) _strtoui64(const char *_String,char **_EndPtr,int _Radix);
  __extension__ __attribute__ ((__dllimport__)) unsigned long long __attribute__((__cdecl__)) _strtoui64_l(const char *_String,char **_EndPtr,int _Radix,_locale_t _Locale);
  ldiv_t __attribute__((__cdecl__)) ldiv(long _Numerator,long _Denominator);
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _ltoa(long _Value,char *_Dest,int _Radix) ;
  int __attribute__((__cdecl__)) mblen(const char *_Ch,size_t _MaxCount);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _mblen_l(const char *_Ch,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _mbstrlen(const char *_Str);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _mbstrlen_l(const char *_Str,_locale_t _Locale);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _mbstrnlen(const char *_Str,size_t _MaxCount);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _mbstrnlen_l(const char *_Str,size_t _MaxCount,_locale_t _Locale);
  int __attribute__((__cdecl__)) mbtowc(wchar_t * __restrict__ _DstCh,const char * __restrict__ _SrcCh,size_t _SrcSizeInBytes);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _mbtowc_l(wchar_t * __restrict__ _DstCh,const char * __restrict__ _SrcCh,size_t _SrcSizeInBytes,_locale_t _Locale);
  size_t __attribute__((__cdecl__)) mbstowcs(wchar_t * __restrict__ _Dest,const char * __restrict__ _Source,size_t _MaxCount);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _mbstowcs_l(wchar_t * __restrict__ _Dest,const char * __restrict__ _Source,size_t _MaxCount,_locale_t _Locale);
  int __attribute__((__cdecl__)) mkstemp(char *template_name);
  int __attribute__((__cdecl__)) rand(void);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _set_error_mode(int _Mode);
  void __attribute__((__cdecl__)) srand(unsigned int _Seed);
# 449 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
inline __attribute__((__cdecl__))
double __attribute__((__cdecl__)) __attribute__ ((__nothrow__)) strtod(const char * __restrict__ _Str,char ** __restrict__ _EndPtr)
{
  double __attribute__((__cdecl__)) __mingw_strtod (const char * __restrict__, char ** __restrict__);
  return __mingw_strtod( _Str, _EndPtr);
}

inline __attribute__((__cdecl__))
float __attribute__((__cdecl__)) __attribute__ ((__nothrow__)) strtof(const char * __restrict__ _Str,char ** __restrict__ _EndPtr)
{
  float __attribute__((__cdecl__)) __mingw_strtof (const char * __restrict__, char ** __restrict__);
  return __mingw_strtof( _Str, _EndPtr);
}






  long double __attribute__((__cdecl__)) __attribute__ ((__nothrow__)) strtold(const char * __restrict__ , char ** __restrict__ );


  extern double __attribute__((__cdecl__)) __attribute__ ((__nothrow__))
  __strtod (const char * __restrict__ , char ** __restrict__);







  float __attribute__((__cdecl__)) __mingw_strtof (const char * __restrict__, char ** __restrict__);
  double __attribute__((__cdecl__)) __mingw_strtod (const char * __restrict__, char ** __restrict__);
  long double __attribute__((__cdecl__)) __mingw_strtold(const char * __restrict__, char ** __restrict__);

  __attribute__ ((__dllimport__)) double __attribute__((__cdecl__)) _strtod_l(const char * __restrict__ _Str,char ** __restrict__ _EndPtr,_locale_t _Locale);
  long __attribute__((__cdecl__)) strtol(const char * __restrict__ _Str,char ** __restrict__ _EndPtr,int _Radix);
  __attribute__ ((__dllimport__)) long __attribute__((__cdecl__)) _strtol_l(const char * __restrict__ _Str,char ** __restrict__ _EndPtr,int _Radix,_locale_t _Locale);
  unsigned long __attribute__((__cdecl__)) strtoul(const char * __restrict__ _Str,char ** __restrict__ _EndPtr,int _Radix);
  __attribute__ ((__dllimport__)) unsigned long __attribute__((__cdecl__)) _strtoul_l(const char * __restrict__ _Str,char ** __restrict__ _EndPtr,int _Radix,_locale_t _Locale);


  int __attribute__((__cdecl__)) system(const char *_Command);

  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _ultoa(unsigned long _Value,char *_Dest,int _Radix) ;
  int __attribute__((__cdecl__)) wctomb(char *_MbCh,wchar_t _WCh) ;
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wctomb_l(char *_MbCh,wchar_t _WCh,_locale_t _Locale) ;
  size_t __attribute__((__cdecl__)) wcstombs(char * __restrict__ _Dest,const wchar_t * __restrict__ _Source,size_t _MaxCount) ;
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _wcstombs_l(char * __restrict__ _Dest,const wchar_t * __restrict__ _Source,size_t _MaxCount,_locale_t _Locale) ;



  void *__attribute__((__cdecl__)) calloc(size_t _NumOfElements,size_t _SizeOfElements);
  void __attribute__((__cdecl__)) free(void *_Memory);
  void *__attribute__((__cdecl__)) malloc(size_t _Size);
  void *__attribute__((__cdecl__)) realloc(void *_Memory,size_t _NewSize);
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _recalloc(void *_Memory,size_t _Count,size_t _Size);
# 517 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _aligned_offset_malloc(size_t _Size,size_t _Alignment,size_t _Offset);
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _aligned_realloc(void *_Memory,size_t _Size,size_t _Alignment);
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _aligned_recalloc(void *_Memory,size_t _Count,size_t _Size,size_t _Alignment);
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _aligned_offset_realloc(void *_Memory,size_t _Size,size_t _Alignment,size_t _Offset);
  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _aligned_offset_recalloc(void *_Memory,size_t _Count,size_t _Size,size_t _Alignment,size_t _Offset);





  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _itow(int _Value,wchar_t *_Dest,int _Radix) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _ltow(long _Value,wchar_t *_Dest,int _Radix) ;
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _ultow(unsigned long _Value,wchar_t *_Dest,int _Radix) ;

  double __attribute__((__cdecl__)) __mingw_wcstod(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr);
  float __attribute__((__cdecl__)) __mingw_wcstof(const wchar_t * __restrict__ nptr, wchar_t ** __restrict__ endptr);
  long double __attribute__((__cdecl__)) __mingw_wcstold(const wchar_t * __restrict__, wchar_t ** __restrict__);


  inline __attribute__((__cdecl__))
  double __attribute__((__cdecl__)) wcstod(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr){
    return __mingw_wcstod(_Str,_EndPtr);
  }
  inline __attribute__((__cdecl__))
  float __attribute__((__cdecl__)) wcstof(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr){
    return __mingw_wcstof(_Str,_EndPtr);
  }






  long double __attribute__((__cdecl__)) wcstold(const wchar_t * __restrict__, wchar_t ** __restrict__);

  __attribute__ ((__dllimport__)) double __attribute__((__cdecl__)) _wcstod_l(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr,_locale_t _Locale);
  long __attribute__((__cdecl__)) wcstol(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr,int _Radix);
  __attribute__ ((__dllimport__)) long __attribute__((__cdecl__)) _wcstol_l(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr,int _Radix,_locale_t _Locale);
  unsigned long __attribute__((__cdecl__)) wcstoul(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr,int _Radix);
  __attribute__ ((__dllimport__)) unsigned long __attribute__((__cdecl__)) _wcstoul_l(const wchar_t * __restrict__ _Str,wchar_t ** __restrict__ _EndPtr,int _Radix,_locale_t _Locale);
  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wgetenv(const wchar_t *_VarName) ;


  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wsystem(const wchar_t *_Command);

  __attribute__ ((__dllimport__)) double __attribute__((__cdecl__)) _wtof(const wchar_t *_Str);
  __attribute__ ((__dllimport__)) double __attribute__((__cdecl__)) _wtof_l(const wchar_t *_Str,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wtoi(const wchar_t *_Str);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wtoi_l(const wchar_t *_Str,_locale_t _Locale);
  __attribute__ ((__dllimport__)) long __attribute__((__cdecl__)) _wtol(const wchar_t *_Str);
  __attribute__ ((__dllimport__)) long __attribute__((__cdecl__)) _wtol_l(const wchar_t *_Str,_locale_t _Locale);

  __extension__ __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _i64tow(long long _Val,wchar_t *_DstBuf,int _Radix) ;
  __extension__ __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _ui64tow(unsigned long long _Val,wchar_t *_DstBuf,int _Radix) ;
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _wtoi64(const wchar_t *_Str);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _wtoi64_l(const wchar_t *_Str,_locale_t _Locale);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _wcstoi64(const wchar_t *_Str,wchar_t **_EndPtr,int _Radix);
  __extension__ __attribute__ ((__dllimport__)) long long __attribute__((__cdecl__)) _wcstoi64_l(const wchar_t *_Str,wchar_t **_EndPtr,int _Radix,_locale_t _Locale);
  __extension__ __attribute__ ((__dllimport__)) unsigned long long __attribute__((__cdecl__)) _wcstoui64(const wchar_t *_Str,wchar_t **_EndPtr,int _Radix);
  __extension__ __attribute__ ((__dllimport__)) unsigned long long __attribute__((__cdecl__)) _wcstoui64_l(const wchar_t *_Str ,wchar_t **_EndPtr,int _Radix,_locale_t _Locale);


  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _putenv(const char *_EnvString);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _wputenv(const wchar_t *_EnvString);



  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _fullpath(char *_FullPath,const char *_Path,size_t _SizeInBytes);
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _ecvt(double _Val,int _NumOfDigits,int *_PtDec,int *_PtSign) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _fcvt(double _Val,int _NumOfDec,int *_PtDec,int *_PtSign) ;
  __attribute__ ((__dllimport__)) char *__attribute__((__cdecl__)) _gcvt(double _Val,int _NumOfDigits,char *_DstBuf) ;
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atodbl(_CRT_DOUBLE *_Result,char *_Str);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atoldbl(_LDOUBLE *_Result,char *_Str);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atoflt(_CRT_FLOAT *_Result,char *_Str);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atodbl_l(_CRT_DOUBLE *_Result,char *_Str,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atoldbl_l(_LDOUBLE *_Result,char *_Str,_locale_t _Locale);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _atoflt_l(_CRT_FLOAT *_Result,char *_Str,_locale_t _Locale);
# 616 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _makepath(char *_Path,const char *_Drive,const char *_Dir,const char *_Filename,const char *_Ext);
  _onexit_t __attribute__((__cdecl__)) _onexit(_onexit_t _Func);



  void __attribute__((__cdecl__)) perror(const char *_ErrMsg);

       
       


  __extension__ unsigned long long __attribute__((__cdecl__)) _rotl64(unsigned long long _Val,int _Shift);
  __extension__ unsigned long long __attribute__((__cdecl__)) _rotr64(unsigned long long Value,int Shift);
       
       
       
       


  unsigned int __attribute__((__cdecl__)) _rotr(unsigned int _Val,int _Shift);
  unsigned int __attribute__((__cdecl__)) _rotl(unsigned int _Val,int _Shift);
       
       
  __extension__ unsigned long long __attribute__((__cdecl__)) _rotr64(unsigned long long _Val,int _Shift);
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _searchenv(const char *_Filename,const char *_EnvVar,char *_ResultPath) ;
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _splitpath(const char *_FullPath,char *_Drive,char *_Dir,char *_Filename,char *_Ext) ;
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _swab(char *_Buf1,char *_Buf2,int _SizeInBytes);



  __attribute__ ((__dllimport__)) wchar_t *__attribute__((__cdecl__)) _wfullpath(wchar_t *_FullPath,const wchar_t *_Path,size_t _SizeInWords);
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _wmakepath(wchar_t *_ResultPath,const wchar_t *_Drive,const wchar_t *_Dir,const wchar_t *_Filename,const wchar_t *_Ext);


  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _wperror(const wchar_t *_ErrMsg);

  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _wsearchenv(const wchar_t *_Filename,const wchar_t *_EnvVar,wchar_t *_ResultPath) ;
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _wsplitpath(const wchar_t *_FullPath,wchar_t *_Drive,wchar_t *_Dir,wchar_t *_Filename,wchar_t *_Ext) ;


  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _beep(unsigned _Frequency,unsigned _Duration) __attribute__ ((__deprecated__));

  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _seterrormode(int _Mode) __attribute__ ((__deprecated__));
  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) _sleep(unsigned long _Duration) __attribute__ ((__deprecated__));
# 680 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
  char *__attribute__((__cdecl__)) ecvt(double _Val,int _NumOfDigits,int *_PtDec,int *_PtSign) ;
  char *__attribute__((__cdecl__)) fcvt(double _Val,int _NumOfDec,int *_PtDec,int *_PtSign) ;
  char *__attribute__((__cdecl__)) gcvt(double _Val,int _NumOfDigits,char *_DstBuf) ;
  char *__attribute__((__cdecl__)) itoa(int _Val,char *_DstBuf,int _Radix) ;
  char *__attribute__((__cdecl__)) ltoa(long _Val,char *_DstBuf,int _Radix) ;
  int __attribute__((__cdecl__)) putenv(const char *_EnvString) ;



  void __attribute__((__cdecl__)) swab(char *_Buf1,char *_Buf2,int _SizeInBytes) ;


  char *__attribute__((__cdecl__)) ultoa(unsigned long _Val,char *_Dstbuf,int _Radix) ;
  _onexit_t __attribute__((__cdecl__)) onexit(_onexit_t _Func);





  typedef struct { __extension__ long long quot, rem; } lldiv_t;

  __extension__ lldiv_t __attribute__((__cdecl__)) lldiv(long long, long long);

  __extension__ long long __attribute__((__cdecl__)) llabs(long long);




  __extension__ long long __attribute__((__cdecl__)) strtoll(const char * __restrict__, char ** __restrict, int);
  __extension__ unsigned long long __attribute__((__cdecl__)) strtoull(const char * __restrict__, char ** __restrict__, int);


  __extension__ long long __attribute__((__cdecl__)) atoll (const char *);
# 735 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 3
}


#pragma pack(pop)

# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/stdlib_s.h" 1 3
# 9 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/stdlib_s.h" 3
# 1 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 1 3
# 30 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 1 3
# 31 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 2 3
# 10 "C:/msys64/mingw64/x86_64-w64-mingw32/include/sec_api/stdlib_s.h" 2 3




extern "C" {


  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _dupenv_s(char **_PBuffer,size_t *_PBufferSizeInBytes,const char *_VarName);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _itoa_s(int _Value,char *_DstBuf,size_t _Size,int _Radix);
  extern "C++" { template <size_t __size> inline errno_t __attribute__((__cdecl__)) _itoa_s(int _Value, char (&_Dest)[__size], int _Radix) { return _itoa_s(_Value, _Dest, __size, _Radix); } }
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _i64toa_s(long long _Val,char *_DstBuf,size_t _Size,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ui64toa_s(unsigned long long _Val,char *_DstBuf,size_t _Size,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ltoa_s(long _Val,char *_DstBuf,size_t _Size,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) mbstowcs_s(size_t *_PtNumOfCharConverted,wchar_t *_DstBuf,size_t _SizeInWords,const char *_SrcBuf,size_t _MaxCount);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _mbstowcs_s_l(size_t *_PtNumOfCharConverted,wchar_t *_DstBuf,size_t _SizeInWords,const char *_SrcBuf,size_t _MaxCount,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ultoa_s(unsigned long _Val,char *_DstBuf,size_t _Size,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wctomb_s_l(int *_SizeConverted,char *_MbCh,size_t _SizeInBytes,wchar_t _WCh,_locale_t _Locale);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) wcstombs_s(size_t *_PtNumOfCharConverted,char *_Dst,size_t _DstSizeInBytes,const wchar_t *_Src,size_t _MaxCountInBytes);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wcstombs_s_l(size_t *_PtNumOfCharConverted,char *_Dst,size_t _DstSizeInBytes,const wchar_t *_Src,size_t _MaxCountInBytes,_locale_t _Locale);



  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _itow_s (int _Val,wchar_t *_DstBuf,size_t _SizeInWords,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ltow_s (long _Val,wchar_t *_DstBuf,size_t _SizeInWords,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ultow_s (unsigned long _Val,wchar_t *_DstBuf,size_t _SizeInWords,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wgetenv_s(size_t *_ReturnSize,wchar_t *_DstBuf,size_t _DstSizeInWords,const wchar_t *_VarName);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wdupenv_s(wchar_t **_Buffer,size_t *_BufferSizeInWords,const wchar_t *_VarName);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _i64tow_s(long long _Val,wchar_t *_DstBuf,size_t _SizeInWords,int _Radix);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ui64tow_s(unsigned long long _Val,wchar_t *_DstBuf,size_t _SizeInWords,int _Radix);



  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _ecvt_s(char *_DstBuf,size_t _Size,double _Val,int _NumOfDights,int *_PtDec,int *_PtSign);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _fcvt_s(char *_DstBuf,size_t _Size,double _Val,int _NumOfDec,int *_PtDec,int *_PtSign);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _gcvt_s(char *_DstBuf,size_t _Size,double _Val,int _NumOfDigits);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _makepath_s(char *_PathResult,size_t _Size,const char *_Drive,const char *_Dir,const char *_Filename,const char *_Ext);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _putenv_s(const char *_Name,const char *_Value);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _searchenv_s(const char *_Filename,const char *_EnvVar,char *_ResultPath,size_t _SizeInBytes);

  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _splitpath_s(const char *_FullPath,char *_Drive,size_t _DriveSize,char *_Dir,size_t _DirSize,char *_Filename,size_t _FilenameSize,char *_Ext,size_t _ExtSize);
  extern "C++" { template <size_t __drive_size, size_t __dir_size, size_t __name_size, size_t __ext_size> inline errno_t __attribute__((__cdecl__)) _splitpath_s(const char *_Dest, char (&__drive)[__drive_size], char (&__dir)[__dir_size], char (&__name)[__name_size], char (&__ext)[__ext_size]) { return _splitpath_s(_Dest, __drive, __drive_size, __dir, __dir_size, __name, __name_size, __ext, __ext_size); } }



  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wmakepath_s(wchar_t *_PathResult,size_t _SizeInWords,const wchar_t *_Drive,const wchar_t *_Dir,const wchar_t *_Filename,const wchar_t *_Ext);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wputenv_s(const wchar_t *_Name,const wchar_t *_Value);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wsearchenv_s(const wchar_t *_Filename,const wchar_t *_EnvVar,wchar_t *_ResultPath,size_t _SizeInWords);

  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _wsplitpath_s(const wchar_t *_FullPath,wchar_t *_Drive,size_t _DriveSizeInWords,wchar_t *_Dir,size_t _DirSizeInWords,wchar_t *_Filename,size_t _FilenameSizeInWords,wchar_t *_Ext,size_t _ExtSizeInWords);
  extern "C++" { template <size_t __drive_size, size_t __dir_size, size_t __name_size, size_t __ext_size> inline errno_t __attribute__((__cdecl__)) _wsplitpath_s(const wchar_t *_Dest, wchar_t (&__drive)[__drive_size], wchar_t (&__dir)[__dir_size], wchar_t (&__name)[__name_size], wchar_t (&__ext)[__ext_size]) { return _wsplitpath_s(_Dest, __drive, __drive_size, __dir, __dir_size, __name, __name_size, __ext, __ext_size); } }





  __attribute__ ((__dllimport__)) void __attribute__((__cdecl__)) qsort_s(void *_Base,size_t _NumOfElements,size_t _SizeOfElements,int (__attribute__((__cdecl__)) *_PtFuncCompare)(void *,const void *,const void *),void *_Context);





}
# 741 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 2 3
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 1 3
# 11 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
#pragma pack(push,_CRT_PACKING)


extern "C" {
# 46 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
  typedef struct _heapinfo {
    int *_pentry;
    size_t _size;
    int _useflag;
  } _HEAPINFO;


  extern unsigned int _amblksiz;
# 103 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
void * __mingw_aligned_malloc (size_t _Size, size_t _Alignment);
void __mingw_aligned_free (void *_Memory);
void * __mingw_aligned_offset_realloc (void *_Memory, size_t _Size, size_t _Alignment, size_t _Offset);
void * __mingw_aligned_realloc (void *_Memory, size_t _Size, size_t _Offset);



  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _resetstkoflw (void);
  __attribute__ ((__dllimport__)) unsigned long __attribute__((__cdecl__)) _set_malloc_crt_max_wait(unsigned long _NewValue);

  __attribute__ ((__dllimport__)) void *__attribute__((__cdecl__)) _expand(void *_Memory,size_t _NewSize);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _msize(void *_Memory);






  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _get_sbh_threshold(void);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _set_sbh_threshold(size_t _NewValue);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _set_amblksiz(size_t _Value);
  __attribute__ ((__dllimport__)) errno_t __attribute__((__cdecl__)) _get_amblksiz(size_t *_Value);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _heapadd(void *_Memory,size_t _Size);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _heapchk(void);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _heapmin(void);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _heapset(unsigned int _Fill);
  __attribute__ ((__dllimport__)) int __attribute__((__cdecl__)) _heapwalk(_HEAPINFO *_EntryInfo);
  __attribute__ ((__dllimport__)) size_t __attribute__((__cdecl__)) _heapused(size_t *_Used,size_t *_Commit);
  __attribute__ ((__dllimport__)) intptr_t __attribute__((__cdecl__)) _get_heap_handle(void);
# 144 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
  static __inline void *_MarkAllocaS(void *_Ptr,unsigned int _Marker) {
    if(_Ptr) {
      *((unsigned int*)_Ptr) = _Marker;
      _Ptr = (char*)_Ptr + 16;
    }
    return _Ptr;
  }
# 163 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
  static __inline void __attribute__((__cdecl__)) _freea(void *_Memory) {
    unsigned int _Marker;
    if(_Memory) {
      _Memory = (char*)_Memory - 16;
      _Marker = *(unsigned int *)_Memory;
      if(_Marker==0xDDDD) {
 free(_Memory);
      }





    }
  }
# 206 "C:/msys64/mingw64/x86_64-w64-mingw32/include/malloc.h" 3
}


#pragma pack(pop)
# 742 "C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h" 2 3
# 76 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 2 3

# 1 "C:/msys64/mingw64/include/c++/7.3.0/bits/std_abs.h" 1 3
# 33 "C:/msys64/mingw64/include/c++/7.3.0/bits/std_abs.h" 3
       
# 34 "C:/msys64/mingw64/include/c++/7.3.0/bits/std_abs.h" 3
# 46 "C:/msys64/mingw64/include/c++/7.3.0/bits/std_abs.h" 3
extern "C++"
{
namespace std
{


  using ::abs;


  inline long
  abs(long __i) { return __builtin_labs(__i); }



  inline long long
  abs(long long __x) { return __builtin_llabs (__x); }







  inline constexpr double
  abs(double __x)
  { return __builtin_fabs(__x); }

  inline constexpr float
  abs(float __x)
  { return __builtin_fabsf(__x); }

  inline constexpr long double
  abs(long double __x)
  { return __builtin_fabsl(__x); }
# 106 "C:/msys64/mingw64/include/c++/7.3.0/bits/std_abs.h" 3

}
}
# 78 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 2 3
# 118 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3
extern "C++"
{
namespace std
{


  using ::div_t;
  using ::ldiv_t;

  using ::abort;
  using ::atexit;





  using ::atof;
  using ::atoi;
  using ::atol;
  using ::bsearch;
  using ::calloc;
  using ::div;
  using ::exit;
  using ::free;
  using ::getenv;
  using ::labs;
  using ::ldiv;
  using ::malloc;

  using ::mblen;
  using ::mbstowcs;
  using ::mbtowc;

  using ::qsort;





  using ::rand;
  using ::realloc;
  using ::srand;
  using ::strtod;
  using ::strtol;
  using ::strtoul;
  using ::system;

  using ::wcstombs;
  using ::wctomb;



  inline ldiv_t
  div(long __i, long __j) { return ldiv(__i, __j); }




}
# 189 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3
namespace __gnu_cxx
{



  using ::lldiv_t;





  using ::_Exit;



  using ::llabs;

  inline lldiv_t
  div(long long __n, long long __d)
  { lldiv_t __q; __q.quot = __n / __d; __q.rem = __n % __d; return __q; }

  using ::lldiv;
# 221 "C:/msys64/mingw64/include/c++/7.3.0/cstdlib" 3
  using ::atoll;
  using ::strtoll;
  using ::strtoull;

  using ::strtof;
  using ::strtold;


}

namespace std
{

  using ::__gnu_cxx::lldiv_t;

  using ::__gnu_cxx::_Exit;

  using ::__gnu_cxx::llabs;
  using ::__gnu_cxx::div;
  using ::__gnu_cxx::lldiv;

  using ::__gnu_cxx::atoll;
  using ::__gnu_cxx::strtof;
  using ::__gnu_cxx::strtoll;
  using ::__gnu_cxx::strtoull;
  using ::__gnu_cxx::strtold;
}



}
# 37 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 2 3 4

using std::abort;
using std::atexit;
using std::exit;
# 51 "C:/msys64/mingw64/include/c++/7.3.0/stdlib.h" 3 4
using std::div_t;
using std::ldiv_t;

using std::abs;
using std::atof;
using std::atoi;
using std::atol;
using std::bsearch;
using std::calloc;
using std::div;
using std::free;
using std::getenv;
using std::labs;
using std::ldiv;
using std::malloc;

using std::mblen;
using std::mbstowcs;
using std::mbtowc;

using std::qsort;
using std::rand;
using std::realloc;
using std::srand;
using std::strtod;
using std::strtol;
using std::strtoul;
using std::system;

using std::wcstombs;
using std::wctomb;
# 28 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mm_malloc.h" 2 3 4
# 1 "C:/msys64/mingw64/x86_64-w64-mingw32/include/errno.h" 1 3 4
# 12 "C:/msys64/mingw64/x86_64-w64-mingw32/include/errno.h" 3 4
extern "C" {
# 239 "C:/msys64/mingw64/x86_64-w64-mingw32/include/errno.h" 3 4
}
# 29 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/mm_malloc.h" 2 3 4

static __inline__ void *
_aligned_malloc(size_t __size,size_t __align)
{
  void * __malloc_ptr;
  void * __aligned_ptr;


  if (__align & (__align - 1))
    {
      (*_errno()) = 22;
      return ((void *) 0);
    }

  if (__size == 0)
    return ((void *) 0);





    if (__align < 2 * sizeof (void *))
      __align = 2 * sizeof (void *);

  __malloc_ptr = malloc (__size + __align);
  if (!__malloc_ptr)
    return ((void *) 0);


  __aligned_ptr = (void *) (((size_t) __malloc_ptr + __align)
       & ~((size_t) (__align) - 1));


  ((void **) __aligned_ptr)[-1] = __malloc_ptr;

  return __aligned_ptr;
}

static __inline__ void
_aligned_free(void *__aligned_ptr)
{
  if (__aligned_ptr)
    free (((void **) __aligned_ptr)[-1]);
}
# 35 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 2 3 4


enum _mm_hint
{

  _MM_HINT_ET0 = 7,
  _MM_HINT_ET1 = 6,
  _MM_HINT_T0 = 3,
  _MM_HINT_T1 = 2,
  _MM_HINT_T2 = 1,
  _MM_HINT_NTA = 0
};
# 68 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern "C" {




typedef float __m128 __attribute__ ((__vector_size__ (16), __may_alias__));


typedef float __m128_u __attribute__ ((__vector_size__ (16), __may_alias__, __aligned__ (1)));


typedef float __v4sf __attribute__ ((__vector_size__ (16)));
# 113 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_undefined_ps (void)
{
  __m128 __Y = __Y;
  return __Y;
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setzero_ps (void)
{
  return __extension__ (__m128){ 0.0f, 0.0f, 0.0f, 0.0f };
}





extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_addss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_subss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_mulss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_div_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_divss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sqrt_ss (__m128 __A)
{
  return (__m128) __builtin_ia32_sqrtss ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp_ss (__m128 __A)
{
  return (__m128) __builtin_ia32_rcpss ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt_ss (__m128 __A)
{
  return (__m128) __builtin_ia32_rsqrtss ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_minss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_maxss ((__v4sf)__A, (__v4sf)__B);
}



extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_ps (__m128 __A, __m128 __B)
{
  return (__m128) ((__v4sf)__A + (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_ps (__m128 __A, __m128 __B)
{
  return (__m128) ((__v4sf)__A - (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_ps (__m128 __A, __m128 __B)
{
  return (__m128) ((__v4sf)__A * (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_div_ps (__m128 __A, __m128 __B)
{
  return (__m128) ((__v4sf)__A / (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sqrt_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_sqrtps ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_rcpps ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_rsqrtps ((__v4sf)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_minps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_maxps ((__v4sf)__A, (__v4sf)__B);
}



extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_and_ps (__m128 __A, __m128 __B)
{
  return __builtin_ia32_andps (__A, __B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_andnot_ps (__m128 __A, __m128 __B)
{
  return __builtin_ia32_andnps (__A, __B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_or_ps (__m128 __A, __m128 __B)
{
  return __builtin_ia32_orps (__A, __B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_xor_ps (__m128 __A, __m128 __B)
{
  return __builtin_ia32_xorps (__A, __B);
}





extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpeqss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpltss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpless ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf) __A,
     (__v4sf)
     __builtin_ia32_cmpltss ((__v4sf) __B,
        (__v4sf)
        __A));
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf) __A,
     (__v4sf)
     __builtin_ia32_cmpless ((__v4sf) __B,
        (__v4sf)
        __A));
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpneqss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnlt_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpnltss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnle_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpnless ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpngt_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf) __A,
     (__v4sf)
     __builtin_ia32_cmpnltss ((__v4sf) __B,
         (__v4sf)
         __A));
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnge_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf) __A,
     (__v4sf)
     __builtin_ia32_cmpnless ((__v4sf) __B,
         (__v4sf)
         __A));
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpord_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpordss ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpunord_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpunordss ((__v4sf)__A, (__v4sf)__B);
}





extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpeqps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpltps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpleps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpgtps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpgeps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpneqps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnlt_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpnltps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnle_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpnleps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpngt_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpngtps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnge_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpngeps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpord_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpordps ((__v4sf)__A, (__v4sf)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpunord_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_cmpunordps ((__v4sf)__A, (__v4sf)__B);
}




extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comieq_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comieq ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comilt_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comilt ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comile_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comile ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comigt_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comigt ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comige_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comige ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comineq_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_comineq ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomieq_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomieq ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomilt_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomilt ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomile_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomile ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomigt_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomigt ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomige_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomige ((__v4sf)__A, (__v4sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomineq_ss (__m128 __A, __m128 __B)
{
  return __builtin_ia32_ucomineq ((__v4sf)__A, (__v4sf)__B);
}



extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si32 (__m128 __A)
{
  return __builtin_ia32_cvtss2si ((__v4sf) __A);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvt_ss2si (__m128 __A)
{
  return _mm_cvtss_si32 (__A);
}






extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si64 (__m128 __A)
{
  return __builtin_ia32_cvtss2si64 ((__v4sf) __A);
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si64x (__m128 __A)
{
  return __builtin_ia32_cvtss2si64 ((__v4sf) __A);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi32 (__m128 __A)
{
  return (__m64) __builtin_ia32_cvtps2pi ((__v4sf) __A);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvt_ps2pi (__m128 __A)
{
  return _mm_cvtps_pi32 (__A);
}


extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si32 (__m128 __A)
{
  return __builtin_ia32_cvttss2si ((__v4sf) __A);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtt_ss2si (__m128 __A)
{
  return _mm_cvttss_si32 (__A);
}





extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si64 (__m128 __A)
{
  return __builtin_ia32_cvttss2si64 ((__v4sf) __A);
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si64x (__m128 __A)
{
  return __builtin_ia32_cvttss2si64 ((__v4sf) __A);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttps_pi32 (__m128 __A)
{
  return (__m64) __builtin_ia32_cvttps2pi ((__v4sf) __A);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtt_ps2pi (__m128 __A)
{
  return _mm_cvttps_pi32 (__A);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi32_ss (__m128 __A, int __B)
{
  return (__m128) __builtin_ia32_cvtsi2ss ((__v4sf) __A, __B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvt_si2ss (__m128 __A, int __B)
{
  return _mm_cvtsi32_ss (__A, __B);
}





extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_ss (__m128 __A, long long __B)
{
  return (__m128) __builtin_ia32_cvtsi642ss ((__v4sf) __A, __B);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64x_ss (__m128 __A, long long __B)
{
  return (__m128) __builtin_ia32_cvtsi642ss ((__v4sf) __A, __B);
}




extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi32_ps (__m128 __A, __m64 __B)
{
  return (__m128) __builtin_ia32_cvtpi2ps ((__v4sf) __A, (__v2si)__B);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvt_pi2ps (__m128 __A, __m64 __B)
{
  return _mm_cvtpi32_ps (__A, __B);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi16_ps (__m64 __A)
{
  __v4hi __sign;
  __v2si __hisi, __losi;
  __v4sf __zero, __ra, __rb;




  __sign = __builtin_ia32_pcmpgtw ((__v4hi)0LL, (__v4hi)__A);


  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __sign);
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __sign);


  __zero = (__v4sf) _mm_setzero_ps ();
  __ra = __builtin_ia32_cvtpi2ps (__zero, __losi);
  __rb = __builtin_ia32_cvtpi2ps (__ra, __hisi);

  return (__m128) __builtin_ia32_movlhps (__ra, __rb);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpu16_ps (__m64 __A)
{
  __v2si __hisi, __losi;
  __v4sf __zero, __ra, __rb;


  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, (__v4hi)0LL);
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, (__v4hi)0LL);


  __zero = (__v4sf) _mm_setzero_ps ();
  __ra = __builtin_ia32_cvtpi2ps (__zero, __losi);
  __rb = __builtin_ia32_cvtpi2ps (__ra, __hisi);

  return (__m128) __builtin_ia32_movlhps (__ra, __rb);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi8_ps (__m64 __A)
{
  __v8qi __sign;




  __sign = __builtin_ia32_pcmpgtb ((__v8qi)0LL, (__v8qi)__A);


  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __sign);

  return _mm_cvtpi16_ps(__A);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpu8_ps(__m64 __A)
{
  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, (__v8qi)0LL);
  return _mm_cvtpu16_ps(__A);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi32x2_ps(__m64 __A, __m64 __B)
{
  __v4sf __zero = (__v4sf) _mm_setzero_ps ();
  __v4sf __sfa = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__A);
  __v4sf __sfb = __builtin_ia32_cvtpi2ps (__sfa, (__v2si)__B);
  return (__m128) __builtin_ia32_movlhps (__sfa, __sfb);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi16(__m128 __A)
{
  __v4sf __hisf = (__v4sf)__A;
  __v4sf __losf = __builtin_ia32_movhlps (__hisf, __hisf);
  __v2si __hisi = __builtin_ia32_cvtps2pi (__hisf);
  __v2si __losi = __builtin_ia32_cvtps2pi (__losf);
  return (__m64) __builtin_ia32_packssdw (__hisi, __losi);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi8(__m128 __A)
{
  __v4hi __tmp = (__v4hi) _mm_cvtps_pi16 (__A);
  return (__m64) __builtin_ia32_packsswb (__tmp, (__v4hi)0LL);
}
# 759 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpckhps ((__v4sf)__A, (__v4sf)__B);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpcklps ((__v4sf)__A, (__v4sf)__B);
}



extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadh_pi (__m128 __A, __m64 const *__P)
{
  return (__m128) __builtin_ia32_loadhps ((__v4sf)__A, (const __v2sf *)__P);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeh_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storehps ((__v2sf *)__P, (__v4sf)__A);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movehl_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movhlps ((__v4sf)__A, (__v4sf)__B);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movelh_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movlhps ((__v4sf)__A, (__v4sf)__B);
}



extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadl_pi (__m128 __A, __m64 const *__P)
{
  return (__m128) __builtin_ia32_loadlps ((__v4sf)__A, (const __v2sf *)__P);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storel_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storelps ((__v2sf *)__P, (__v4sf)__A);
}


extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_ps (__m128 __A)
{
  return __builtin_ia32_movmskps ((__v4sf)__A);
}


extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_getcsr (void)
{
  return __builtin_ia32_stmxcsr ();
}


extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_GET_EXCEPTION_STATE (void)
{
  return _mm_getcsr() & 0x003f;
}

extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_GET_EXCEPTION_MASK (void)
{
  return _mm_getcsr() & 0x1f80;
}

extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_GET_ROUNDING_MODE (void)
{
  return _mm_getcsr() & 0x6000;
}

extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_GET_FLUSH_ZERO_MODE (void)
{
  return _mm_getcsr() & 0x8000;
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setcsr (unsigned int __I)
{
  __builtin_ia32_ldmxcsr (__I);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_SET_EXCEPTION_STATE(unsigned int __mask)
{
  _mm_setcsr((_mm_getcsr() & ~0x003f) | __mask);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_SET_EXCEPTION_MASK (unsigned int __mask)
{
  _mm_setcsr((_mm_getcsr() & ~0x1f80) | __mask);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_SET_ROUNDING_MODE (unsigned int __mode)
{
  _mm_setcsr((_mm_getcsr() & ~0x6000) | __mode);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_SET_FLUSH_ZERO_MODE (unsigned int __mode)
{
  _mm_setcsr((_mm_getcsr() & ~0x8000) | __mode);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_ss (float __F)
{
  return __extension__ (__m128)(__v4sf){ __F, 0.0f, 0.0f, 0.0f };
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_ps (float __F)
{
  return __extension__ (__m128)(__v4sf){ __F, __F, __F, __F };
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_ps1 (float __F)
{
  return _mm_set1_ps (__F);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_ss (float const *__P)
{
  return _mm_set_ss (*__P);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load1_ps (float const *__P)
{
  return _mm_set1_ps (*__P);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_ps1 (float const *__P)
{
  return _mm_load1_ps (__P);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_ps (float const *__P)
{
  return *(__m128 *)__P;
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadu_ps (float const *__P)
{
  return *(__m128_u *)__P;
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadr_ps (float const *__P)
{
  __v4sf __tmp = *(__v4sf *)__P;
  return (__m128) __builtin_ia32_shufps (__tmp, __tmp, (((0) << 6) | ((1) << 4) | ((2) << 2) | (3)));
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_ps (const float __Z, const float __Y, const float __X, const float __W)
{
  return __extension__ (__m128)(__v4sf){ __W, __X, __Y, __Z };
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_ps (float __Z, float __Y, float __X, float __W)
{
  return __extension__ (__m128)(__v4sf){ __Z, __Y, __X, __W };
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_ss (float *__P, __m128 __A)
{
  *__P = ((__v4sf)__A)[0];
}

extern __inline float __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_f32 (__m128 __A)
{
  return ((__v4sf)__A)[0];
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_ps (float *__P, __m128 __A)
{
  *(__m128 *)__P = __A;
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeu_ps (float *__P, __m128 __A)
{
  *(__m128_u *)__P = __A;
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store1_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, (((0) << 6) | ((0) << 4) | ((0) << 2) | (0)));
  _mm_storeu_ps (__P, __tmp);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_ps1 (float *__P, __m128 __A)
{
  _mm_store1_ps (__P, __A);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storer_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, (((0) << 6) | ((1) << 4) | ((2) << 2) | (3)));
  _mm_store_ps (__P, __tmp);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_move_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movss ((__v4sf)__A, (__v4sf)__B);
}
# 1064 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxsw ((__v4hi)__A, (__v4hi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmaxsw (__m64 __A, __m64 __B)
{
  return _mm_max_pi16 (__A, __B);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxub ((__v8qi)__A, (__v8qi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmaxub (__m64 __A, __m64 __B)
{
  return _mm_max_pu8 (__A, __B);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminsw ((__v4hi)__A, (__v4hi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pminsw (__m64 __A, __m64 __B)
{
  return _mm_min_pi16 (__A, __B);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminub ((__v8qi)__A, (__v8qi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pminub (__m64 __A, __m64 __B)
{
  return _mm_min_pu8 (__A, __B);
}


extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_pi8 (__m64 __A)
{
  return __builtin_ia32_pmovmskb ((__v8qi)__A);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmovmskb (__m64 __A)
{
  return _mm_movemask_pi8 (__A);
}



extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhi_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmulhuw ((__v4hi)__A, (__v4hi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pmulhuw (__m64 __A, __m64 __B)
{
  return _mm_mulhi_pu16 (__A, __B);
}
# 1166 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskmove_si64 (__m64 __A, __m64 __N, char *__P)
{
  __builtin_ia32_maskmovq ((__v8qi)__A, (__v8qi)__N, __P);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_maskmovq (__m64 __A, __m64 __N, char *__P)
{
  _mm_maskmove_si64 (__A, __N, __P);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pavgb ((__v8qi)__A, (__v8qi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pavgb (__m64 __A, __m64 __B)
{
  return _mm_avg_pu8 (__A, __B);
}


extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pavgw ((__v4hi)__A, (__v4hi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pavgw (__m64 __A, __m64 __B)
{
  return _mm_avg_pu16 (__A, __B);
}




extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sad_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_psadbw ((__v8qi)__A, (__v8qi)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_psadbw (__m64 __A, __m64 __B)
{
  return _mm_sad_pu8 (__A, __B);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_pi (__m64 *__P, __m64 __A)
{
  __builtin_ia32_movntq ((unsigned long long *)__P, (unsigned long long)__A);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_ps (float *__P, __m128 __A)
{
  __builtin_ia32_movntps (__P, (__v4sf)__A);
}



extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sfence (void)
{
  __builtin_ia32_sfence ();
}
# 1256 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
}



# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 1 3 4
# 31 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 2 3 4
# 40 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
typedef double __v2df __attribute__ ((__vector_size__ (16)));
typedef long long __v2di __attribute__ ((__vector_size__ (16)));
typedef unsigned long long __v2du __attribute__ ((__vector_size__ (16)));
typedef int __v4si __attribute__ ((__vector_size__ (16)));
typedef unsigned int __v4su __attribute__ ((__vector_size__ (16)));
typedef short __v8hi __attribute__ ((__vector_size__ (16)));
typedef unsigned short __v8hu __attribute__ ((__vector_size__ (16)));
typedef char __v16qi __attribute__ ((__vector_size__ (16)));
typedef unsigned char __v16qu __attribute__ ((__vector_size__ (16)));



typedef long long __m128i __attribute__ ((__vector_size__ (16), __may_alias__));
typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));


typedef long long __m128i_u __attribute__ ((__vector_size__ (16), __may_alias__, __aligned__ (1)));
typedef double __m128d_u __attribute__ ((__vector_size__ (16), __may_alias__, __aligned__ (1)));






extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_sd (double __F)
{
  return __extension__ (__m128d){ __F, 0.0 };
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_pd (double __F)
{
  return __extension__ (__m128d){ __F, __F };
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pd1 (double __F)
{
  return _mm_set1_pd (__F);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_pd (double __W, double __X)
{
  return __extension__ (__m128d){ __X, __W };
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_pd (double __W, double __X)
{
  return __extension__ (__m128d){ __W, __X };
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_undefined_pd (void)
{
  __m128d __Y = __Y;
  return __Y;
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setzero_pd (void)
{
  return __extension__ (__m128d){ 0.0, 0.0 };
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_move_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_pd (double const *__P)
{
  return *(__m128d *)__P;
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadu_pd (double const *__P)
{
  return *(__m128d_u *)__P;
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load1_pd (double const *__P)
{
  return _mm_set1_pd (*__P);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_sd (double const *__P)
{
  return _mm_set_sd (*__P);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_pd1 (double const *__P)
{
  return _mm_load1_pd (__P);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadr_pd (double const *__P)
{
  __m128d __tmp = _mm_load_pd (__P);
  return __builtin_ia32_shufpd (__tmp, __tmp, (((0) << 1) | (1)));
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_pd (double *__P, __m128d __A)
{
  *(__m128d *)__P = __A;
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeu_pd (double *__P, __m128d __A)
{
  *(__m128d_u *)__P = __A;
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_sd (double *__P, __m128d __A)
{
  *__P = ((__v2df)__A)[0];
}

extern __inline double __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_f64 (__m128d __A)
{
  return ((__v2df)__A)[0];
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storel_pd (double *__P, __m128d __A)
{
  _mm_store_sd (__P, __A);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeh_pd (double *__P, __m128d __A)
{
  *__P = ((__v2df)__A)[1];
}



extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store1_pd (double *__P, __m128d __A)
{
  _mm_store_pd (__P, __builtin_ia32_shufpd (__A, __A, (((0) << 1) | (0))));
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_pd1 (double *__P, __m128d __A)
{
  _mm_store1_pd (__P, __A);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storer_pd (double *__P, __m128d __A)
{
  _mm_store_pd (__P, __builtin_ia32_shufpd (__A, __A, (((0) << 1) | (1))));
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi128_si32 (__m128i __A)
{
  return __builtin_ia32_vec_ext_v4si ((__v4si)__A, 0);
}



extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi128_si64 (__m128i __A)
{
  return ((__v2di)__A)[0];
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi128_si64x (__m128i __A)
{
  return ((__v2di)__A)[0];
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_pd (__m128d __A, __m128d __B)
{
  return (__m128d) ((__v2df)__A + (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_addsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_pd (__m128d __A, __m128d __B)
{
  return (__m128d) ((__v2df)__A - (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_subsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_pd (__m128d __A, __m128d __B)
{
  return (__m128d) ((__v2df)__A * (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_mulsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_div_pd (__m128d __A, __m128d __B)
{
  return (__m128d) ((__v2df)__A / (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_div_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_divsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sqrt_pd (__m128d __A)
{
  return (__m128d)__builtin_ia32_sqrtpd ((__v2df)__A);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sqrt_sd (__m128d __A, __m128d __B)
{
  __v2df __tmp = __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
  return (__m128d)__builtin_ia32_sqrtsd ((__v2df)__tmp);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_minpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_minsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_maxpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_maxsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_and_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_andpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_andnot_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_andnpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_or_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_orpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_xor_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_xorpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpeqpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpltpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmplepd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpgtpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpgepd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpneqpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnlt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnltpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnle_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnlepd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpngt_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpngtpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnge_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpngepd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpord_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpordpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpunord_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpunordpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpeqsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpltsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmplesd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
      (__v2df)
      __builtin_ia32_cmpltsd ((__v2df) __B,
         (__v2df)
         __A));
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
      (__v2df)
      __builtin_ia32_cmplesd ((__v2df) __B,
         (__v2df)
         __A));
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpneqsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnlt_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnltsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnle_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpnlesd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpngt_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
      (__v2df)
      __builtin_ia32_cmpnltsd ((__v2df) __B,
          (__v2df)
          __A));
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpnge_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_movsd ((__v2df) __A,
      (__v2df)
      __builtin_ia32_cmpnlesd ((__v2df) __B,
          (__v2df)
          __A));
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpord_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpordsd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpunord_sd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_cmpunordsd ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comieq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdeq ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comilt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdlt ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comile_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdle ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comigt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdgt ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comige_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdge ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_comineq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_comisdneq ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomieq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdeq ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomilt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdlt ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomile_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdle ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomigt_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdgt ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomige_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdge ((__v2df)__A, (__v2df)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_ucomineq_sd (__m128d __A, __m128d __B)
{
  return __builtin_ia32_ucomisdneq ((__v2df)__A, (__v2df)__B);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_epi64x (long long __q1, long long __q0)
{
  return __extension__ (__m128i)(__v2di){ __q0, __q1 };
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_epi64 (__m64 __q1, __m64 __q0)
{
  return _mm_set_epi64x ((long long)__q1, (long long)__q0);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_epi32 (int __q3, int __q2, int __q1, int __q0)
{
  return __extension__ (__m128i)(__v4si){ __q0, __q1, __q2, __q3 };
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_epi16 (short __q7, short __q6, short __q5, short __q4,
        short __q3, short __q2, short __q1, short __q0)
{
  return __extension__ (__m128i)(__v8hi){
    __q0, __q1, __q2, __q3, __q4, __q5, __q6, __q7 };
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_epi8 (char __q15, char __q14, char __q13, char __q12,
       char __q11, char __q10, char __q09, char __q08,
       char __q07, char __q06, char __q05, char __q04,
       char __q03, char __q02, char __q01, char __q00)
{
  return __extension__ (__m128i)(__v16qi){
    __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07,
    __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15
  };
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_epi64x (long long __A)
{
  return _mm_set_epi64x (__A, __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_epi64 (__m64 __A)
{
  return _mm_set_epi64 (__A, __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_epi32 (int __A)
{
  return _mm_set_epi32 (__A, __A, __A, __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_epi16 (short __A)
{
  return _mm_set_epi16 (__A, __A, __A, __A, __A, __A, __A, __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_epi8 (char __A)
{
  return _mm_set_epi8 (__A, __A, __A, __A, __A, __A, __A, __A,
         __A, __A, __A, __A, __A, __A, __A, __A);
}




extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_epi64 (__m64 __q0, __m64 __q1)
{
  return _mm_set_epi64 (__q1, __q0);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_epi32 (int __q0, int __q1, int __q2, int __q3)
{
  return _mm_set_epi32 (__q3, __q2, __q1, __q0);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_epi16 (short __q0, short __q1, short __q2, short __q3,
         short __q4, short __q5, short __q6, short __q7)
{
  return _mm_set_epi16 (__q7, __q6, __q5, __q4, __q3, __q2, __q1, __q0);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_epi8 (char __q00, char __q01, char __q02, char __q03,
        char __q04, char __q05, char __q06, char __q07,
        char __q08, char __q09, char __q10, char __q11,
        char __q12, char __q13, char __q14, char __q15)
{
  return _mm_set_epi8 (__q15, __q14, __q13, __q12, __q11, __q10, __q09, __q08,
         __q07, __q06, __q05, __q04, __q03, __q02, __q01, __q00);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_si128 (__m128i const *__P)
{
  return *__P;
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadu_si128 (__m128i_u const *__P)
{
  return *__P;
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadl_epi64 (__m128i_u const *__P)
{
  return _mm_set_epi64 ((__m64)0LL, *(__m64_u *)__P);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_si128 (__m128i *__P, __m128i __B)
{
  *__P = __B;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeu_si128 (__m128i_u *__P, __m128i __B)
{
  *__P = __B;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storel_epi64 (__m128i_u *__P, __m128i __B)
{
  *(__m64_u *)__P = (__m64) ((__v2di)__B)[0];
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movepi64_pi64 (__m128i __B)
{
  return (__m64) ((__v2di)__B)[0];
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movpi64_epi64 (__m64 __A)
{
  return _mm_set_epi64 ((__m64)0LL, __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_move_epi64 (__m128i __A)
{
  return (__m128i)__builtin_ia32_movq128 ((__v2di) __A);
}


extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_undefined_si128 (void)
{
  __m128i __Y = __Y;
  return __Y;
}


extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setzero_si128 (void)
{
  return __extension__ (__m128i)(__v4si){ 0, 0, 0, 0 };
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi32_pd (__m128i __A)
{
  return (__m128d)__builtin_ia32_cvtdq2pd ((__v4si) __A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi32_ps (__m128i __A)
{
  return (__m128)__builtin_ia32_cvtdq2ps ((__v4si) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpd_epi32 (__m128d __A)
{
  return (__m128i)__builtin_ia32_cvtpd2dq ((__v2df) __A);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpd_pi32 (__m128d __A)
{
  return (__m64)__builtin_ia32_cvtpd2pi ((__v2df) __A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpd_ps (__m128d __A)
{
  return (__m128)__builtin_ia32_cvtpd2ps ((__v2df) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttpd_epi32 (__m128d __A)
{
  return (__m128i)__builtin_ia32_cvttpd2dq ((__v2df) __A);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttpd_pi32 (__m128d __A)
{
  return (__m64)__builtin_ia32_cvttpd2pi ((__v2df) __A);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi32_pd (__m64 __A)
{
  return (__m128d)__builtin_ia32_cvtpi2pd ((__v2si) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_epi32 (__m128 __A)
{
  return (__m128i)__builtin_ia32_cvtps2dq ((__v4sf) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttps_epi32 (__m128 __A)
{
  return (__m128i)__builtin_ia32_cvttps2dq ((__v4sf) __A);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pd (__m128 __A)
{
  return (__m128d)__builtin_ia32_cvtps2pd ((__v4sf) __A);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_si32 (__m128d __A)
{
  return __builtin_ia32_cvtsd2si ((__v2df) __A);
}



extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_si64 (__m128d __A)
{
  return __builtin_ia32_cvtsd2si64 ((__v2df) __A);
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_si64x (__m128d __A)
{
  return __builtin_ia32_cvtsd2si64 ((__v2df) __A);
}


extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_si32 (__m128d __A)
{
  return __builtin_ia32_cvttsd2si ((__v2df) __A);
}



extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_si64 (__m128d __A)
{
  return __builtin_ia32_cvttsd2si64 ((__v2df) __A);
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_si64x (__m128d __A)
{
  return __builtin_ia32_cvttsd2si64 ((__v2df) __A);
}


extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_ss (__m128 __A, __m128d __B)
{
  return (__m128)__builtin_ia32_cvtsd2ss ((__v4sf) __A, (__v2df) __B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi32_sd (__m128d __A, int __B)
{
  return (__m128d)__builtin_ia32_cvtsi2sd ((__v2df) __A, __B);
}



extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_sd (__m128d __A, long long __B)
{
  return (__m128d)__builtin_ia32_cvtsi642sd ((__v2df) __A, __B);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64x_sd (__m128d __A, long long __B)
{
  return (__m128d)__builtin_ia32_cvtsi642sd ((__v2df) __A, __B);
}


extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_sd (__m128d __A, __m128 __B)
{
  return (__m128d)__builtin_ia32_cvtss2sd ((__v2df) __A, (__v4sf)__B);
}
# 918 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_unpckhpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_pd (__m128d __A, __m128d __B)
{
  return (__m128d)__builtin_ia32_unpcklpd ((__v2df)__A, (__v2df)__B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadh_pd (__m128d __A, double const *__B)
{
  return (__m128d)__builtin_ia32_loadhpd ((__v2df)__A, __B);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadl_pd (__m128d __A, double const *__B)
{
  return (__m128d)__builtin_ia32_loadlpd ((__v2df)__A, __B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_pd (__m128d __A)
{
  return __builtin_ia32_movmskpd ((__v2df)__A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packs_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packsswb128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packs_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packssdw128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packus_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_packuswb128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhbw128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhwd128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhdq128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckhqdq128 ((__v2di)__A, (__v2di)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklbw128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklwd128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpckldq128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_punpcklqdq128 ((__v2di)__A, (__v2di)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v16qu)__A + (__v16qu)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hu)__A + (__v8hu)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v4su)__A + (__v4su)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_add_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v2du)__A + (__v2du)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddsb128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddsw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddusb128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_adds_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_paddusw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v16qu)__A - (__v16qu)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hu)__A - (__v8hu)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v4su)__A - (__v4su)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sub_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v2du)__A - (__v2du)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubsb128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubsw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubusb128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_subs_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psubusw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_madd_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaddwd128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhi_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmulhw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mullo_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hu)__A * (__v8hu)__B);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_su32 (__m64 __A, __m64 __B)
{
  return (__m64)__builtin_ia32_pmuludq ((__v2si)__A, (__v2si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_epu32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmuludq128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psllwi128 ((__v8hi)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_pslldi128 ((__v4si)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_slli_epi64 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psllqi128 ((__v2di)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srai_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrawi128 ((__v8hi)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srai_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psradi128 ((__v4si)__A, __B);
}
# 1205 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_epi16 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrlwi128 ((__v8hi)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_epi32 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrldi128 ((__v4si)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srli_epi64 (__m128i __A, int __B)
{
  return (__m128i)__builtin_ia32_psrlqi128 ((__v2di)__A, __B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psllw128((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pslld128((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psllq128((__v2di)__A, (__v2di)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sra_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psraw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sra_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psrad128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psrlw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psrld128 ((__v4si)__A, (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psrlq128 ((__v2di)__A, (__v2di)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_and_si128 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v2du)__A & (__v2du)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_andnot_si128 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pandn128 ((__v2di)__A, (__v2di)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_or_si128 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v2du)__A | (__v2du)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_xor_si128 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v2du)__A ^ (__v2du)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v16qi)__A == (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hi)__A == (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v4si)__A == (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v16qi)__A < (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hi)__A < (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v4si)__A < (__v4si)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi8 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v16qi)__A > (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v8hi)__A > (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) ((__v4si)__A > (__v4si)__B);
}
# 1369 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaxsw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmaxub128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epi16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pminsw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pminub128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_epi8 (__m128i __A)
{
  return __builtin_ia32_pmovmskb128 ((__v16qi)__A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhi_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pmulhuw128 ((__v8hi)__A, (__v8hi)__B);
}
# 1432 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/emmintrin.h" 3 4
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskmoveu_si128 (__m128i __A, __m128i __B, char *__C)
{
  __builtin_ia32_maskmovdqu ((__v16qi)__A, (__v16qi)__B, __C);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pavgb128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_epu16 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_pavgw128 ((__v8hi)__A, (__v8hi)__B);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sad_epu8 (__m128i __A, __m128i __B)
{
  return (__m128i)__builtin_ia32_psadbw128 ((__v16qi)__A, (__v16qi)__B);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_si32 (int *__A, int __B)
{
  __builtin_ia32_movnti (__A, __B);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_si64 (long long int *__A, long long int __B)
{
  __builtin_ia32_movnti64 (__A, __B);
}


extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_si128 (__m128i *__A, __m128i __B)
{
  __builtin_ia32_movntdq ((__v2di *)__A, (__v2di)__B);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_pd (double *__A, __m128d __B)
{
  __builtin_ia32_movntpd (__A, (__v2df)__B);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_clflush (void const *__A)
{
  __builtin_ia32_clflush (__A);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_lfence (void)
{
  __builtin_ia32_lfence ();
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mfence (void)
{
  __builtin_ia32_mfence ();
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi32_si128 (int __A)
{
  return _mm_set_epi32 (0, 0, 0, __A);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64_si128 (long long __A)
{
  return _mm_set_epi64x (0, __A);
}


extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64x_si128 (long long __A)
{
  return _mm_set_epi64x (0, __A);
}




extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castpd_ps(__m128d __A)
{
  return (__m128) __A;
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castpd_si128(__m128d __A)
{
  return (__m128i) __A;
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castps_pd(__m128 __A)
{
  return (__m128d) __A;
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castps_si128(__m128 __A)
{
  return (__m128i) __A;
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castsi128_ps(__m128i __A)
{
  return (__m128) __A;
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_castsi128_pd(__m128i __A)
{
  return (__m128d) __A;
}
# 1261 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 2 3 4
# 1272 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/xmmintrin.h" 3 4
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_pause (void)
{
  __builtin_ia32_pause ();
}
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4



# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/pmmintrin.h" 1 3 4
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/pmmintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("sse3")
# 49 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/pmmintrin.h" 3 4
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_addsub_ps (__m128 __X, __m128 __Y)
{
  return (__m128) __builtin_ia32_addsubps ((__v4sf)__X, (__v4sf)__Y);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_ps (__m128 __X, __m128 __Y)
{
  return (__m128) __builtin_ia32_haddps ((__v4sf)__X, (__v4sf)__Y);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_ps (__m128 __X, __m128 __Y)
{
  return (__m128) __builtin_ia32_hsubps ((__v4sf)__X, (__v4sf)__Y);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movehdup_ps (__m128 __X)
{
  return (__m128) __builtin_ia32_movshdup ((__v4sf)__X);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_moveldup_ps (__m128 __X)
{
  return (__m128) __builtin_ia32_movsldup ((__v4sf)__X);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_addsub_pd (__m128d __X, __m128d __Y)
{
  return (__m128d) __builtin_ia32_addsubpd ((__v2df)__X, (__v2df)__Y);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_pd (__m128d __X, __m128d __Y)
{
  return (__m128d) __builtin_ia32_haddpd ((__v2df)__X, (__v2df)__Y);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_pd (__m128d __X, __m128d __Y)
{
  return (__m128d) __builtin_ia32_hsubpd ((__v2df)__X, (__v2df)__Y);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loaddup_pd (double const *__P)
{
  return _mm_load1_pd (__P);
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movedup_pd (__m128d __X)
{
  return ((__m128d)__builtin_ia32_shufpd ((__v2df)(__m128d)(__X), (__v2df)(__m128d)(__X), (int)((((0) << 1) | (0)))));
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_lddqu_si128 (__m128i const *__P)
{
  return (__m128i) __builtin_ia32_lddqu ((char const *)__P);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_monitor (void const * __P, unsigned int __E, unsigned int __H)
{
  __builtin_ia32_monitor (__P, __E, __H);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mwait (unsigned int __E, unsigned int __H)
{
  __builtin_ia32_mwait (__E, __H);
}



#pragma GCC pop_options
# 38 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/tmmintrin.h" 1 3 4
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/tmmintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("ssse3")



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phaddw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phaddd128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadds_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phaddsw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phaddw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadd_pi32 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phaddd ((__v2si)__X, (__v2si)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hadds_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phaddsw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phsubw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phsubd128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsubs_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_phsubsw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phsubw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsub_pi32 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phsubd ((__v2si)__X, (__v2si)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_hsubs_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_phsubsw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maddubs_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmaddubsw128 ((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maddubs_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_pmaddubsw ((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhrs_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmulhrsw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhrs_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_pmulhrsw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_epi8 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pshufb128 ((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_pi8 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_pshufb ((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_epi8 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psignb128 ((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_epi16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psignw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psignd128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_pi8 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_psignb ((__v8qi)__X, (__v8qi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_pi16 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_psignw ((__v4hi)__X, (__v4hi)__Y);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sign_pi32 (__m64 __X, __m64 __Y)
{
  return (__m64) __builtin_ia32_psignd ((__v2si)__X, (__v2si)__Y);
}
# 208 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/tmmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_epi8 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pabsb128 ((__v16qi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_epi16 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pabsw128 ((__v8hi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pabsd128 ((__v4si)__X);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_pi8 (__m64 __X)
{
  return (__m64) __builtin_ia32_pabsb ((__v8qi)__X);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_pi16 (__m64 __X)
{
  return (__m64) __builtin_ia32_pabsw ((__v4hi)__X);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_pi32 (__m64 __X)
{
  return (__m64) __builtin_ia32_pabsd ((__v2si)__X);
}



#pragma GCC pop_options
# 40 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ammintrin.h" 1 3 4
# 34 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ammintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("sse4a")



extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_sd (double * __P, __m128d __Y)
{
  __builtin_ia32_movntsd (__P, (__v2df) __Y);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_ss (float * __P, __m128 __Y)
{
  __builtin_ia32_movntss (__P, (__v4sf) __Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_extract_si64 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_extrq ((__v2di) __X, (__v16qi) __Y);
}
# 69 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ammintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_insert_si64 (__m128i __X,__m128i __Y)
{
  return (__m128i) __builtin_ia32_insertq ((__v2di)__X, (__v2di)__Y);
}
# 90 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/ammintrin.h" 3 4
#pragma GCC pop_options
# 42 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 1 3 4
# 35 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("sse4.1")
# 66 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testz_si128 (__m128i __M, __m128i __V)
{
  return __builtin_ia32_ptestz128 ((__v2di)__M, (__v2di)__V);
}



extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testc_si128 (__m128i __M, __m128i __V)
{
  return __builtin_ia32_ptestc128 ((__v2di)__M, (__v2di)__V);
}



extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testnzc_si128 (__m128i __M, __m128i __V)
{
  return __builtin_ia32_ptestnzc128 ((__v2di)__M, (__v2di)__V);
}
# 178 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_blendv_epi8 (__m128i __X, __m128i __Y, __m128i __M)
{
  return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__X,
            (__v16qi)__Y,
            (__v16qi)__M);
}
# 203 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_blendv_ps (__m128 __X, __m128 __Y, __m128 __M)
{
  return (__m128) __builtin_ia32_blendvps ((__v4sf)__X,
        (__v4sf)__Y,
        (__v4sf)__M);
}
# 228 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_blendv_pd (__m128d __X, __m128d __Y, __m128d __M)
{
  return (__m128d) __builtin_ia32_blendvpd ((__v2df)__X,
         (__v2df)__Y,
         (__v2df)__M);
}
# 267 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi64 (__m128i __X, __m128i __Y)
{
  return (__m128i) ((__v2di)__X == (__v2di)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epi8 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pminsb128 ((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epi8 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi)__X, (__v16qi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epu16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pminuw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epu16 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi)__X, (__v8hi)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pminsd128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epu32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pminud128 ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epu32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmaxud128 ((__v4si)__X, (__v4si)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mullo_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) ((__v4su)__X * (__v4su)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mul_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__X, (__v4si)__Y);
}
# 473 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_minpos_epu16 (__m128i __X)
{
  return (__m128i) __builtin_ia32_phminposuw128 ((__v8hi)__X);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi8_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxbd128 ((__v16qi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi16_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxwd128 ((__v8hi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi8_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxbq128 ((__v16qi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi32_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxdq128 ((__v4si)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi16_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxwq128 ((__v8hi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi8_epi16 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxbw128 ((__v16qi)__X);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu8_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxbd128 ((__v16qi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu16_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxwd128 ((__v8hi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu8_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxbq128 ((__v16qi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu32_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxdq128 ((__v4si)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu16_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxwq128 ((__v8hi)__X);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu8_epi16 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxbw128 ((__v16qi)__X);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_packus_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_packusdw128 ((__v4si)__X, (__v4si)__Y);
}
# 581 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_load_si128 (__m128i *__X)
{
  return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __X);
}


#pragma GCC push_options
#pragma GCC target("sse4.2")
# 795 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 3 4
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi64 (__m128i __X, __m128i __Y)
{
  return (__m128i) ((__v2di)__X > (__v2di)__Y);
}



#pragma GCC pop_options




#pragma GCC pop_options


# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/popcntintrin.h" 1 3 4
# 28 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/popcntintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("popcnt")




extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_popcnt_u32 (unsigned int __X)
{
  return __builtin_popcount (__X);
}


extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_popcnt_u64 (unsigned long long __X)
{
  return __builtin_popcountll (__X);
}




#pragma GCC pop_options
# 812 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/smmintrin.h" 2 3 4


#pragma GCC push_options
#pragma GCC target("sse4.1")




#pragma GCC push_options
#pragma GCC target("sse4.2")




extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_crc32_u8 (unsigned int __C, unsigned char __V)
{
  return __builtin_ia32_crc32qi (__C, __V);
}

extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_crc32_u16 (unsigned int __C, unsigned short __V)
{
  return __builtin_ia32_crc32hi (__C, __V);
}

extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_crc32_u32 (unsigned int __C, unsigned int __V)
{
  return __builtin_ia32_crc32si (__C, __V);
}


extern __inline unsigned long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_crc32_u64 (unsigned long long __C, unsigned long long __V)
{
  return __builtin_ia32_crc32di (__C, __V);
}




#pragma GCC pop_options




#pragma GCC pop_options
# 44 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/wmmintrin.h" 1 3 4
# 36 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/wmmintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("aes")





extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_aesdec_si128 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_aesdec128 ((__v2di)__X, (__v2di)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_aesdeclast_si128 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_aesdeclast128 ((__v2di)__X,
       (__v2di)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_aesenc_si128 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_aesenc128 ((__v2di)__X, (__v2di)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_aesenclast_si128 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_aesenclast128 ((__v2di)__X, (__v2di)__Y);
}



extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_aesimc_si128 (__m128i __X)
{
  return (__m128i) __builtin_ia32_aesimc128 ((__v2di)__X);
}
# 99 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/wmmintrin.h" 3 4
#pragma GCC pop_options





#pragma GCC push_options
#pragma GCC target("pclmul")
# 129 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/wmmintrin.h" 3 4
#pragma GCC pop_options
# 46 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/x86intrin.h" 2 3 4


# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 1 3 4
# 41 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 3 4
# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 1 3 4
# 35 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx")




typedef double __v4df __attribute__ ((__vector_size__ (32)));
typedef float __v8sf __attribute__ ((__vector_size__ (32)));
typedef long long __v4di __attribute__ ((__vector_size__ (32)));
typedef unsigned long long __v4du __attribute__ ((__vector_size__ (32)));
typedef int __v8si __attribute__ ((__vector_size__ (32)));
typedef unsigned int __v8su __attribute__ ((__vector_size__ (32)));
typedef short __v16hi __attribute__ ((__vector_size__ (32)));
typedef unsigned short __v16hu __attribute__ ((__vector_size__ (32)));
typedef char __v32qi __attribute__ ((__vector_size__ (32)));
typedef unsigned char __v32qu __attribute__ ((__vector_size__ (32)));



typedef float __m256 __attribute__ ((__vector_size__ (32),
         __may_alias__));
typedef long long __m256i __attribute__ ((__vector_size__ (32),
       __may_alias__));
typedef double __m256d __attribute__ ((__vector_size__ (32),
           __may_alias__));


typedef float __m256_u __attribute__ ((__vector_size__ (32),
           __may_alias__,
           __aligned__ (1)));
typedef long long __m256i_u __attribute__ ((__vector_size__ (32),
         __may_alias__,
         __aligned__ (1)));
typedef double __m256d_u __attribute__ ((__vector_size__ (32),
      __may_alias__,
      __aligned__ (1)));
# 139 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_pd (__m256d __A, __m256d __B)
{
  return (__m256d) ((__v4df)__A + (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_ps (__m256 __A, __m256 __B)
{
  return (__m256) ((__v8sf)__A + (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_addsub_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_addsubpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_addsub_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_addsubps256 ((__v8sf)__A, (__v8sf)__B);
}


extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_and_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_andpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_and_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_andps256 ((__v8sf)__A, (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_andnot_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_andnpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_andnot_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_andnps256 ((__v8sf)__A, (__v8sf)__B);
}
# 217 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_blendv_pd (__m256d __X, __m256d __Y, __m256d __M)
{
  return (__m256d) __builtin_ia32_blendvpd256 ((__v4df)__X,
            (__v4df)__Y,
            (__v4df)__M);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_blendv_ps (__m256 __X, __m256 __Y, __m256 __M)
{
  return (__m256) __builtin_ia32_blendvps256 ((__v8sf)__X,
           (__v8sf)__Y,
           (__v8sf)__M);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_div_pd (__m256d __A, __m256d __B)
{
  return (__m256d) ((__v4df)__A / (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_div_ps (__m256 __A, __m256 __B)
{
  return (__m256) ((__v8sf)__A / (__v8sf)__B);
}
# 262 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hadd_pd (__m256d __X, __m256d __Y)
{
  return (__m256d) __builtin_ia32_haddpd256 ((__v4df)__X, (__v4df)__Y);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hadd_ps (__m256 __X, __m256 __Y)
{
  return (__m256) __builtin_ia32_haddps256 ((__v8sf)__X, (__v8sf)__Y);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hsub_pd (__m256d __X, __m256d __Y)
{
  return (__m256d) __builtin_ia32_hsubpd256 ((__v4df)__X, (__v4df)__Y);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hsub_ps (__m256 __X, __m256 __Y)
{
  return (__m256) __builtin_ia32_hsubps256 ((__v8sf)__X, (__v8sf)__Y);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_maxpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_maxps256 ((__v8sf)__A, (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_minpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_minps256 ((__v8sf)__A, (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mul_pd (__m256d __A, __m256d __B)
{
  return (__m256d) ((__v4df)__A * (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mul_ps (__m256 __A, __m256 __B)
{
  return (__m256) ((__v8sf)__A * (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_or_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_orpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_or_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_orps256 ((__v8sf)__A, (__v8sf)__B);
}
# 358 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_pd (__m256d __A, __m256d __B)
{
  return (__m256d) ((__v4df)__A - (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_ps (__m256 __A, __m256 __B)
{
  return (__m256) ((__v8sf)__A - (__v8sf)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_xor_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_xorpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_xor_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_xorps256 ((__v8sf)__A, (__v8sf)__B);
}
# 446 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi32_pd (__m128i __A)
{
  return (__m256d)__builtin_ia32_cvtdq2pd256 ((__v4si) __A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi32_ps (__m256i __A)
{
  return (__m256)__builtin_ia32_cvtdq2ps256 ((__v8si) __A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtpd_ps (__m256d __A)
{
  return (__m128)__builtin_ia32_cvtpd2ps256 ((__v4df) __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtps_epi32 (__m256 __A)
{
  return (__m256i)__builtin_ia32_cvtps2dq256 ((__v8sf) __A);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtps_pd (__m128 __A)
{
  return (__m256d)__builtin_ia32_cvtps2pd256 ((__v4sf) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvttpd_epi32 (__m256d __A)
{
  return (__m128i)__builtin_ia32_cvttpd2dq256 ((__v4df) __A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtpd_epi32 (__m256d __A)
{
  return (__m128i)__builtin_ia32_cvtpd2dq256 ((__v4df) __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvttps_epi32 (__m256 __A)
{
  return (__m256i)__builtin_ia32_cvttps2dq256 ((__v8sf) __A);
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsd_f64 (__m256d __A)
{
  return __A[0];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtss_f32 (__m256 __A)
{
  return __A[0];
}
# 600 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_zeroall (void)
{
  __builtin_ia32_vzeroall ();
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_zeroupper (void)
{
  __builtin_ia32_vzeroupper ();
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutevar_pd (__m128d __A, __m128i __C)
{
  return (__m128d) __builtin_ia32_vpermilvarpd ((__v2df)__A,
      (__v2di)__C);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutevar_pd (__m256d __A, __m256i __C)
{
  return (__m256d) __builtin_ia32_vpermilvarpd256 ((__v4df)__A,
         (__v4di)__C);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutevar_ps (__m128 __A, __m128i __C)
{
  return (__m128) __builtin_ia32_vpermilvarps ((__v4sf)__A,
            (__v4si)__C);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutevar_ps (__m256 __A, __m256i __C)
{
  return (__m256) __builtin_ia32_vpermilvarps256 ((__v8sf)__A,
        (__v8si)__C);
}
# 719 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcast_ss (float const *__X)
{
  return (__m128) __builtin_ia32_vbroadcastss (__X);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_sd (double const *__X)
{
  return (__m256d) __builtin_ia32_vbroadcastsd256 (__X);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_ss (float const *__X)
{
  return (__m256) __builtin_ia32_vbroadcastss256 (__X);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_pd (__m128d const *__X)
{
  return (__m256d) __builtin_ia32_vbroadcastf128_pd256 (__X);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_ps (__m128 const *__X)
{
  return (__m256) __builtin_ia32_vbroadcastf128_ps256 (__X);
}
# 858 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_load_pd (double const *__P)
{
  return *(__m256d *)__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_store_pd (double *__P, __m256d __A)
{
  *(__m256d *)__P = __A;
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_load_ps (float const *__P)
{
  return *(__m256 *)__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_store_ps (float *__P, __m256 __A)
{
  *(__m256 *)__P = __A;
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_loadu_pd (double const *__P)
{
  return *(__m256d_u *)__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_storeu_pd (double *__P, __m256d __A)
{
  *(__m256d_u *)__P = __A;
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_loadu_ps (float const *__P)
{
  return *(__m256_u *)__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_storeu_ps (float *__P, __m256 __A)
{
  *(__m256_u *)__P = __A;
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_load_si256 (__m256i const *__P)
{
  return *__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_store_si256 (__m256i *__P, __m256i __A)
{
  *__P = __A;
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_loadu_si256 (__m256i_u const *__P)
{
  return *__P;
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_storeu_si256 (__m256i_u *__P, __m256i __A)
{
  *__P = __A;
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskload_pd (double const *__P, __m128i __M)
{
  return (__m128d) __builtin_ia32_maskloadpd ((const __v2df *)__P,
           (__v2di)__M);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskstore_pd (double *__P, __m128i __M, __m128d __A)
{
  __builtin_ia32_maskstorepd ((__v2df *)__P, (__v2di)__M, (__v2df)__A);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskload_pd (double const *__P, __m256i __M)
{
  return (__m256d) __builtin_ia32_maskloadpd256 ((const __v4df *)__P,
       (__v4di)__M);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskstore_pd (double *__P, __m256i __M, __m256d __A)
{
  __builtin_ia32_maskstorepd256 ((__v4df *)__P, (__v4di)__M, (__v4df)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskload_ps (float const *__P, __m128i __M)
{
  return (__m128) __builtin_ia32_maskloadps ((const __v4sf *)__P,
          (__v4si)__M);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskstore_ps (float *__P, __m128i __M, __m128 __A)
{
  __builtin_ia32_maskstoreps ((__v4sf *)__P, (__v4si)__M, (__v4sf)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskload_ps (float const *__P, __m256i __M)
{
  return (__m256) __builtin_ia32_maskloadps256 ((const __v8sf *)__P,
      (__v8si)__M);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskstore_ps (float *__P, __m256i __M, __m256 __A)
{
  __builtin_ia32_maskstoreps256 ((__v8sf *)__P, (__v8si)__M, (__v8sf)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_movehdup_ps (__m256 __X)
{
  return (__m256) __builtin_ia32_movshdup256 ((__v8sf)__X);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_moveldup_ps (__m256 __X)
{
  return (__m256) __builtin_ia32_movsldup256 ((__v8sf)__X);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_movedup_pd (__m256d __X)
{
  return (__m256d) __builtin_ia32_movddup256 ((__v4df)__X);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_lddqu_si256 (__m256i const *__P)
{
  return (__m256i) __builtin_ia32_lddqu256 ((char const *)__P);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_stream_si256 (__m256i *__A, __m256i __B)
{
  __builtin_ia32_movntdq256 ((__v4di *)__A, (__v4di)__B);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_stream_pd (double *__A, __m256d __B)
{
  __builtin_ia32_movntpd256 (__A, (__v4df)__B);
}

extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_stream_ps (float *__P, __m256 __A)
{
  __builtin_ia32_movntps256 (__P, (__v8sf)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rcp_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_rcpps256 ((__v8sf)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rsqrt_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_rsqrtps256 ((__v8sf)__A);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sqrt_pd (__m256d __A)
{
  return (__m256d) __builtin_ia32_sqrtpd256 ((__v4df)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sqrt_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_sqrtps256 ((__v8sf)__A);
}
# 1073 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avxintrin.h" 3 4
extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_unpckhpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_unpcklpd256 ((__v4df)__A, (__v4df)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_unpckhps256 ((__v8sf)__A, (__v8sf)__B);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_unpcklps256 ((__v8sf)__A, (__v8sf)__B);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testz_pd (__m128d __M, __m128d __V)
{
  return __builtin_ia32_vtestzpd ((__v2df)__M, (__v2df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testc_pd (__m128d __M, __m128d __V)
{
  return __builtin_ia32_vtestcpd ((__v2df)__M, (__v2df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testnzc_pd (__m128d __M, __m128d __V)
{
  return __builtin_ia32_vtestnzcpd ((__v2df)__M, (__v2df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testz_ps (__m128 __M, __m128 __V)
{
  return __builtin_ia32_vtestzps ((__v4sf)__M, (__v4sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testc_ps (__m128 __M, __m128 __V)
{
  return __builtin_ia32_vtestcps ((__v4sf)__M, (__v4sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_testnzc_ps (__m128 __M, __m128 __V)
{
  return __builtin_ia32_vtestnzcps ((__v4sf)__M, (__v4sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testz_pd (__m256d __M, __m256d __V)
{
  return __builtin_ia32_vtestzpd256 ((__v4df)__M, (__v4df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testc_pd (__m256d __M, __m256d __V)
{
  return __builtin_ia32_vtestcpd256 ((__v4df)__M, (__v4df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testnzc_pd (__m256d __M, __m256d __V)
{
  return __builtin_ia32_vtestnzcpd256 ((__v4df)__M, (__v4df)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testz_ps (__m256 __M, __m256 __V)
{
  return __builtin_ia32_vtestzps256 ((__v8sf)__M, (__v8sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testc_ps (__m256 __M, __m256 __V)
{
  return __builtin_ia32_vtestcps256 ((__v8sf)__M, (__v8sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testnzc_ps (__m256 __M, __m256 __V)
{
  return __builtin_ia32_vtestnzcps256 ((__v8sf)__M, (__v8sf)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testz_si256 (__m256i __M, __m256i __V)
{
  return __builtin_ia32_ptestz256 ((__v4di)__M, (__v4di)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testc_si256 (__m256i __M, __m256i __V)
{
  return __builtin_ia32_ptestc256 ((__v4di)__M, (__v4di)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testnzc_si256 (__m256i __M, __m256i __V)
{
  return __builtin_ia32_ptestnzc256 ((__v4di)__M, (__v4di)__V);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_movemask_pd (__m256d __A)
{
  return __builtin_ia32_movmskpd256 ((__v4df)__A);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_movemask_ps (__m256 __A)
{
  return __builtin_ia32_movmskps256 ((__v8sf)__A);
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_undefined_pd (void)
{
  __m256d __Y = __Y;
  return __Y;
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_undefined_ps (void)
{
  __m256 __Y = __Y;
  return __Y;
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_undefined_si256 (void)
{
  __m256i __Y = __Y;
  return __Y;
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setzero_pd (void)
{
  return __extension__ (__m256d){ 0.0, 0.0, 0.0, 0.0 };
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setzero_ps (void)
{
  return __extension__ (__m256){ 0.0, 0.0, 0.0, 0.0,
     0.0, 0.0, 0.0, 0.0 };
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setzero_si256 (void)
{
  return __extension__ (__m256i)(__v4di){ 0, 0, 0, 0 };
}


extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_pd (double __A, double __B, double __C, double __D)
{
  return __extension__ (__m256d){ __D, __C, __B, __A };
}


extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_ps (float __A, float __B, float __C, float __D,
        float __E, float __F, float __G, float __H)
{
  return __extension__ (__m256){ __H, __G, __F, __E,
     __D, __C, __B, __A };
}


extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_epi32 (int __A, int __B, int __C, int __D,
    int __E, int __F, int __G, int __H)
{
  return __extension__ (__m256i)(__v8si){ __H, __G, __F, __E,
       __D, __C, __B, __A };
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_epi16 (short __q15, short __q14, short __q13, short __q12,
    short __q11, short __q10, short __q09, short __q08,
    short __q07, short __q06, short __q05, short __q04,
    short __q03, short __q02, short __q01, short __q00)
{
  return __extension__ (__m256i)(__v16hi){
    __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07,
    __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15
  };
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_epi8 (char __q31, char __q30, char __q29, char __q28,
    char __q27, char __q26, char __q25, char __q24,
    char __q23, char __q22, char __q21, char __q20,
    char __q19, char __q18, char __q17, char __q16,
    char __q15, char __q14, char __q13, char __q12,
    char __q11, char __q10, char __q09, char __q08,
    char __q07, char __q06, char __q05, char __q04,
    char __q03, char __q02, char __q01, char __q00)
{
  return __extension__ (__m256i)(__v32qi){
    __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07,
    __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15,
    __q16, __q17, __q18, __q19, __q20, __q21, __q22, __q23,
    __q24, __q25, __q26, __q27, __q28, __q29, __q30, __q31
  };
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set_epi64x (long long __A, long long __B, long long __C,
     long long __D)
{
  return __extension__ (__m256i)(__v4di){ __D, __C, __B, __A };
}


extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_pd (double __A)
{
  return __extension__ (__m256d){ __A, __A, __A, __A };
}


extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_ps (float __A)
{
  return __extension__ (__m256){ __A, __A, __A, __A,
     __A, __A, __A, __A };
}


extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_epi32 (int __A)
{
  return __extension__ (__m256i)(__v8si){ __A, __A, __A, __A,
       __A, __A, __A, __A };
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_epi16 (short __A)
{
  return _mm256_set_epi16 (__A, __A, __A, __A, __A, __A, __A, __A,
      __A, __A, __A, __A, __A, __A, __A, __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_epi8 (char __A)
{
  return _mm256_set_epi8 (__A, __A, __A, __A, __A, __A, __A, __A,
     __A, __A, __A, __A, __A, __A, __A, __A,
     __A, __A, __A, __A, __A, __A, __A, __A,
     __A, __A, __A, __A, __A, __A, __A, __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_set1_epi64x (long long __A)
{
  return __extension__ (__m256i)(__v4di){ __A, __A, __A, __A };
}




extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_pd (double __A, double __B, double __C, double __D)
{
  return _mm256_set_pd (__D, __C, __B, __A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_ps (float __A, float __B, float __C, float __D,
  float __E, float __F, float __G, float __H)
{
  return _mm256_set_ps (__H, __G, __F, __E, __D, __C, __B, __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_epi32 (int __A, int __B, int __C, int __D,
     int __E, int __F, int __G, int __H)
{
  return _mm256_set_epi32 (__H, __G, __F, __E, __D, __C, __B, __A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_epi16 (short __q15, short __q14, short __q13, short __q12,
     short __q11, short __q10, short __q09, short __q08,
     short __q07, short __q06, short __q05, short __q04,
     short __q03, short __q02, short __q01, short __q00)
{
  return _mm256_set_epi16 (__q00, __q01, __q02, __q03,
      __q04, __q05, __q06, __q07,
      __q08, __q09, __q10, __q11,
      __q12, __q13, __q14, __q15);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_epi8 (char __q31, char __q30, char __q29, char __q28,
     char __q27, char __q26, char __q25, char __q24,
     char __q23, char __q22, char __q21, char __q20,
     char __q19, char __q18, char __q17, char __q16,
     char __q15, char __q14, char __q13, char __q12,
     char __q11, char __q10, char __q09, char __q08,
     char __q07, char __q06, char __q05, char __q04,
     char __q03, char __q02, char __q01, char __q00)
{
  return _mm256_set_epi8 (__q00, __q01, __q02, __q03,
     __q04, __q05, __q06, __q07,
     __q08, __q09, __q10, __q11,
     __q12, __q13, __q14, __q15,
     __q16, __q17, __q18, __q19,
     __q20, __q21, __q22, __q23,
     __q24, __q25, __q26, __q27,
     __q28, __q29, __q30, __q31);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_setr_epi64x (long long __A, long long __B, long long __C,
      long long __D)
{
  return _mm256_set_epi64x (__D, __C, __B, __A);
}



extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castpd_ps (__m256d __A)
{
  return (__m256) __A;
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castpd_si256 (__m256d __A)
{
  return (__m256i) __A;
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castps_pd (__m256 __A)
{
  return (__m256d) __A;
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castps_si256(__m256 __A)
{
  return (__m256i) __A;
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castsi256_ps (__m256i __A)
{
  return (__m256) __A;
}

extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castsi256_pd (__m256i __A)
{
  return (__m256d) __A;
}

extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castpd256_pd128 (__m256d __A)
{
  return (__m128d) __builtin_ia32_pd_pd256 ((__v4df)__A);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castps256_ps128 (__m256 __A)
{
  return (__m128) __builtin_ia32_ps_ps256 ((__v8sf)__A);
}

extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castsi256_si128 (__m256i __A)
{
  return (__m128i) __builtin_ia32_si_si256 ((__v8si)__A);
}






extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castpd128_pd256 (__m128d __A)
{
  return (__m256d) __builtin_ia32_pd256_pd ((__v2df)__A);
}

extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castps128_ps256 (__m128 __A)
{
  return (__m256) __builtin_ia32_ps256_ps ((__v4sf)__A);
}

extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm256_castsi128_si256 (__m128i __A)
{
  return (__m256i) __builtin_ia32_si256_si ((__v4si)__A);
}



#pragma GCC pop_options
# 42 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx2")
# 54 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_abs_epi8 (__m256i __A)
{
  return (__m256i)__builtin_ia32_pabsb256 ((__v32qi)__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_abs_epi16 (__m256i __A)
{
  return (__m256i)__builtin_ia32_pabsw256 ((__v16hi)__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_abs_epi32 (__m256i __A)
{
  return (__m256i)__builtin_ia32_pabsd256 ((__v8si)__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_packs_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_packssdw256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_packs_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_packsswb256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_packus_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_packusdw256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_packus_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_packuswb256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v32qu)__A + (__v32qu)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v16hu)__A + (__v16hu)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v8su)__A + (__v8su)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_add_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4du)__A + (__v4du)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_adds_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_paddsb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_adds_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_paddsw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_adds_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_paddusb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_adds_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_paddusw256 ((__v16hi)__A, (__v16hi)__B);
}
# 177 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_and_si256 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4du)__A & (__v4du)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_andnot_si256 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_andnotsi256 ((__v4di)__A, (__v4di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_avg_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pavgb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_avg_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pavgw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_blendv_epi8 (__m256i __X, __m256i __Y, __m256i __M)
{
  return (__m256i) __builtin_ia32_pblendvb256 ((__v32qi)__X,
            (__v32qi)__Y,
            (__v32qi)__M);
}
# 229 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v32qi)__A == (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v16hi)__A == (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v8si)__A == (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4di)__A == (__v4di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v32qi)__A > (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v16hi)__A > (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v8si)__A > (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4di)__A > (__v4di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hadd_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phaddw256 ((__v16hi)__X,
          (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hadd_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phaddd256 ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hadds_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phaddsw256 ((__v16hi)__X,
           (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hsub_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phsubw256 ((__v16hi)__X,
          (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hsub_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phsubd256 ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_hsubs_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_phsubsw256 ((__v16hi)__X,
           (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maddubs_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmaddubsw256 ((__v32qi)__X,
      (__v32qi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_madd_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaddwd256 ((__v16hi)__A,
          (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxsb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxsw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxsd256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxub256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxuw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epu32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmaxud256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminsb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminsw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminsd256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminub256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminuw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epu32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pminud256 ((__v8si)__A, (__v8si)__B);
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_movemask_epi8 (__m256i __A)
{
  return __builtin_ia32_pmovmskb256 ((__v32qi)__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi8_epi16 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxbw256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi8_epi32 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxbd256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi8_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxbq256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi16_epi32 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxwd256 ((__v8hi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi16_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxwq256 ((__v8hi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi32_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxdq256 ((__v4si)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu8_epi16 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxbw256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu8_epi32 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxbd256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu8_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxbq256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu16_epi32 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxwd256 ((__v8hi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu16_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxwq256 ((__v8hi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu32_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxdq256 ((__v4si)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mul_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmuldq256 ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mulhrs_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmulhrsw256 ((__v16hi)__X,
            (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mulhi_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmulhuw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mulhi_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmulhw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mullo_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v16hu)__A * (__v16hu)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mullo_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v8su)__A * (__v8su)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mul_epu32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_pmuludq256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_or_si256 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4du)__A | (__v4du)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sad_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_psadbw256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_shuffle_epi8 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pshufb256 ((__v32qi)__X,
          (__v32qi)__Y);
}
# 624 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sign_epi8 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psignb256 ((__v32qi)__X, (__v32qi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sign_epi16 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psignw256 ((__v16hi)__X, (__v16hi)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sign_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psignd256 ((__v8si)__X, (__v8si)__Y);
}
# 666 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_slli_epi16 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psllwi256 ((__v16hi)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sll_epi16 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psllw256((__v16hi)__A, (__v8hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_slli_epi32 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_pslldi256 ((__v8si)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sll_epi32 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_pslld256((__v8si)__A, (__v4si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_slli_epi64 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psllqi256 ((__v4di)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sll_epi64 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psllq256((__v4di)__A, (__v2di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srai_epi16 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psrawi256 ((__v16hi)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sra_epi16 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psraw256 ((__v16hi)__A, (__v8hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srai_epi32 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psradi256 ((__v8si)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sra_epi32 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psrad256 ((__v8si)__A, (__v4si)__B);
}
# 757 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srli_epi16 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psrlwi256 ((__v16hi)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srl_epi16 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psrlw256((__v16hi)__A, (__v8hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srli_epi32 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psrldi256 ((__v8si)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srl_epi32 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psrld256((__v8si)__A, (__v4si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srli_epi64 (__m256i __A, int __B)
{
  return (__m256i)__builtin_ia32_psrlqi256 ((__v4di)__A, __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srl_epi64 (__m256i __A, __m128i __B)
{
  return (__m256i)__builtin_ia32_psrlq256((__v4di)__A, (__v2di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v32qu)__A - (__v32qu)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v16hu)__A - (__v16hu)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v8su)__A - (__v8su)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sub_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4du)__A - (__v4du)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_subs_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_psubsb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_subs_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_psubsw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_subs_epu8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_psubusb256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_subs_epu16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_psubusw256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpckhbw256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpckhwd256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpckhdq256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpackhi_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpckhqdq256 ((__v4di)__A, (__v4di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_epi8 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpcklbw256 ((__v32qi)__A, (__v32qi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_epi16 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpcklwd256 ((__v16hi)__A, (__v16hi)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpckldq256 ((__v8si)__A, (__v8si)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_unpacklo_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i)__builtin_ia32_punpcklqdq256 ((__v4di)__A, (__v4di)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_xor_si256 (__m256i __A, __m256i __B)
{
  return (__m256i) ((__v4du)__A ^ (__v4du)__B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_stream_load_si256 (__m256i const *__X)
{
  return (__m256i) __builtin_ia32_movntdqa256 ((__v4di *) __X);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastss_ps (__m128 __X)
{
  return (__m128) __builtin_ia32_vbroadcastss_ps ((__v4sf)__X);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastss_ps (__m128 __X)
{
  return (__m256) __builtin_ia32_vbroadcastss_ps256 ((__v4sf)__X);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastsd_pd (__m128d __X)
{
  return (__m256d) __builtin_ia32_vbroadcastsd_pd256 ((__v2df)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastsi128_si256 (__m128i __X)
{
  return (__m256i) __builtin_ia32_vbroadcastsi256 ((__v2di)__X);
}
# 983 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastb_epi8 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pbroadcastb256 ((__v16qi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastw_epi16 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pbroadcastw256 ((__v8hi)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastd_epi32 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pbroadcastd256 ((__v4si)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastq_epi64 (__m128i __X)
{
  return (__m256i) __builtin_ia32_pbroadcastq256 ((__v2di)__X);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastb_epi8 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pbroadcastb128 ((__v16qi)__X);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastw_epi16 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pbroadcastw128 ((__v8hi)__X);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastd_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pbroadcastd128 ((__v4si)__X);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastq_epi64 (__m128i __X)
{
  return (__m128i) __builtin_ia32_pbroadcastq128 ((__v2di)__X);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutevar8x32_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_permvarsi256 ((__v8si)__X, (__v8si)__Y);
}
# 1058 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutevar8x32_ps (__m256 __X, __m256i __Y)
{
  return (__m256) __builtin_ia32_permvarsf256 ((__v8sf)__X, (__v8si)__Y);
}
# 1116 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskload_epi32 (int const *__X, __m256i __M )
{
  return (__m256i) __builtin_ia32_maskloadd256 ((const __v8si *)__X,
      (__v8si)__M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskload_epi64 (long long const *__X, __m256i __M )
{
  return (__m256i) __builtin_ia32_maskloadq256 ((const __v4di *)__X,
      (__v4di)__M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskload_epi32 (int const *__X, __m128i __M )
{
  return (__m128i) __builtin_ia32_maskloadd ((const __v4si *)__X,
          (__v4si)__M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskload_epi64 (long long const *__X, __m128i __M )
{
  return (__m128i) __builtin_ia32_maskloadq ((const __v2di *)__X,
          (__v2di)__M);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskstore_epi32 (int *__X, __m256i __M, __m256i __Y )
{
  __builtin_ia32_maskstored256 ((__v8si *)__X, (__v8si)__M, (__v8si)__Y);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskstore_epi64 (long long *__X, __m256i __M, __m256i __Y )
{
  __builtin_ia32_maskstoreq256 ((__v4di *)__X, (__v4di)__M, (__v4di)__Y);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskstore_epi32 (int *__X, __m128i __M, __m128i __Y )
{
  __builtin_ia32_maskstored ((__v4si *)__X, (__v4si)__M, (__v4si)__Y);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskstore_epi64 (long long *__X, __m128i __M, __m128i __Y )
{
  __builtin_ia32_maskstoreq (( __v2di *)__X, (__v2di)__M, (__v2di)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sllv_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv8si ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_sllv_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv4si ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sllv_epi64 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv4di ((__v4di)__X, (__v4di)__Y);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_sllv_epi64 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv2di ((__v2di)__X, (__v2di)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srav_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrav8si ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_srav_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrav4si ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srlv_epi32 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv8si ((__v8si)__X, (__v8si)__Y);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_srlv_epi32 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv4si ((__v4si)__X, (__v4si)__Y);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srlv_epi64 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv4di ((__v4di)__X, (__v4di)__Y);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_srlv_epi64 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv2di ((__v2di)__X, (__v2di)__Y);
}
# 1905 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx2intrin.h" 3 4
#pragma GCC pop_options
# 44 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512f")




typedef double __v8df __attribute__ ((__vector_size__ (64)));
typedef float __v16sf __attribute__ ((__vector_size__ (64)));
typedef long long __v8di __attribute__ ((__vector_size__ (64)));
typedef unsigned long long __v8du __attribute__ ((__vector_size__ (64)));
typedef int __v16si __attribute__ ((__vector_size__ (64)));
typedef unsigned int __v16su __attribute__ ((__vector_size__ (64)));
typedef short __v32hi __attribute__ ((__vector_size__ (64)));
typedef unsigned short __v32hu __attribute__ ((__vector_size__ (64)));
typedef char __v64qi __attribute__ ((__vector_size__ (64)));
typedef unsigned char __v64qu __attribute__ ((__vector_size__ (64)));



typedef float __m512 __attribute__ ((__vector_size__ (64), __may_alias__));
typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));


typedef float __m512_u __attribute__ ((__vector_size__ (64), __may_alias__, __aligned__ (1)));
typedef long long __m512i_u __attribute__ ((__vector_size__ (64), __may_alias__, __aligned__ (1)));
typedef double __m512d_u __attribute__ ((__vector_size__ (64), __may_alias__, __aligned__ (1)));

typedef unsigned char __mmask8;
typedef unsigned short __mmask16;

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_int2mask (int __M)
{
  return (__mmask16) __M;
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2int (__mmask16 __M)
{
  return (int) __M;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set_epi64 (long long __A, long long __B, long long __C,
    long long __D, long long __E, long long __F,
    long long __G, long long __H)
{
  return __extension__ (__m512i) (__v8di)
  { __H, __G, __F, __E, __D, __C, __B, __A };
}


extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set_epi32 (int __A, int __B, int __C, int __D,
    int __E, int __F, int __G, int __H,
    int __I, int __J, int __K, int __L,
    int __M, int __N, int __O, int __P)
{
  return __extension__ (__m512i)(__v16si)
  { __P, __O, __N, __M, __L, __K, __J, __I,
    __H, __G, __F, __E, __D, __C, __B, __A };
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set_pd (double __A, double __B, double __C, double __D,
        double __E, double __F, double __G, double __H)
{
  return __extension__ (__m512d)
  { __H, __G, __F, __E, __D, __C, __B, __A };
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set_ps (float __A, float __B, float __C, float __D,
        float __E, float __F, float __G, float __H,
        float __I, float __J, float __K, float __L,
        float __M, float __N, float __O, float __P)
{
  return __extension__ (__m512)
  { __P, __O, __N, __M, __L, __K, __J, __I,
    __H, __G, __F, __E, __D, __C, __B, __A };
}
# 134 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_undefined_ps (void)
{
  __m512 __Y = __Y;
  return __Y;
}



extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_undefined_pd (void)
{
  __m512d __Y = __Y;
  return __Y;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_undefined_epi32 (void)
{
  __m512i __Y = __Y;
  return __Y;
}



extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_epi8 (char __A)
{
  return __extension__ (__m512i)(__v64qi)
  { __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A };
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_epi16 (short __A)
{
  return __extension__ (__m512i)(__v32hi)
  { __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A,
    __A, __A, __A, __A, __A, __A, __A, __A };
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_pd (double __A)
{
  return (__m512d) __builtin_ia32_broadcastsd512 (__extension__
        (__v2df) { __A, },
        (__v8df)
        _mm512_undefined_pd (),
        (__mmask8) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_ps (float __A)
{
  return (__m512) __builtin_ia32_broadcastss512 (__extension__
       (__v4sf) { __A, },
       (__v16sf)
       _mm512_undefined_ps (),
       (__mmask16) -1);
}


extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set4_epi32 (int __A, int __B, int __C, int __D)
{
  return __extension__ (__m512i)(__v16si)
  { __D, __C, __B, __A, __D, __C, __B, __A,
    __D, __C, __B, __A, __D, __C, __B, __A };
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set4_epi64 (long long __A, long long __B, long long __C,
     long long __D)
{
  return __extension__ (__m512i) (__v8di)
  { __D, __C, __B, __A, __D, __C, __B, __A };
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set4_pd (double __A, double __B, double __C, double __D)
{
  return __extension__ (__m512d)
  { __D, __C, __B, __A, __D, __C, __B, __A };
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set4_ps (float __A, float __B, float __C, float __D)
{
  return __extension__ (__m512)
  { __D, __C, __B, __A, __D, __C, __B, __A,
    __D, __C, __B, __A, __D, __C, __B, __A };
}
# 258 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_setzero_ps (void)
{
  return __extension__ (__m512){ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
     0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 };
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_setzero_pd (void)
{
  return __extension__ (__m512d) { 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 };
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_setzero_epi32 (void)
{
  return __extension__ (__m512i)(__v8di){ 0, 0, 0, 0, 0, 0, 0, 0 };
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_setzero_si512 (void)
{
  return __extension__ (__m512i)(__v8di){ 0, 0, 0, 0, 0, 0, 0, 0 };
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_movapd512_mask ((__v8df) __A,
        (__v8df) __W,
        (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_movapd512_mask ((__v8df) __A,
        (__v8df)
        _mm512_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movaps512_mask ((__v16sf) __A,
       (__v16sf) __W,
       (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movaps512_mask ((__v16sf) __A,
       (__v16sf)
       _mm512_setzero_ps (),
       (__mmask16) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_load_pd (void const *__P)
{
  return *(__m512d *) __P;
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_load_pd (__m512d __W, __mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_loadapd512_mask ((const __v8df *) __P,
         (__v8df) __W,
         (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_load_pd (__mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_loadapd512_mask ((const __v8df *) __P,
         (__v8df)
         _mm512_setzero_pd (),
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_store_pd (void *__P, __m512d __A)
{
  *(__m512d *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_store_pd (void *__P, __mmask8 __U, __m512d __A)
{
  __builtin_ia32_storeapd512_mask ((__v8df *) __P, (__v8df) __A,
       (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_load_ps (void const *__P)
{
  return *(__m512 *) __P;
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_load_ps (__m512 __W, __mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_loadaps512_mask ((const __v16sf *) __P,
        (__v16sf) __W,
        (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_load_ps (__mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_loadaps512_mask ((const __v16sf *) __P,
        (__v16sf)
        _mm512_setzero_ps (),
        (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_store_ps (void *__P, __m512 __A)
{
  *(__m512 *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_store_ps (void *__P, __mmask16 __U, __m512 __A)
{
  __builtin_ia32_storeaps512_mask ((__v16sf *) __P, (__v16sf) __A,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdqa64_512_mask ((__v8di) __A,
           (__v8di) __W,
           (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdqa64_512_mask ((__v8di) __A,
           (__v8di)
           _mm512_setzero_si512 (),
           (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_load_epi64 (void const *__P)
{
  return *(__m512i *) __P;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_load_epi64 (__m512i __W, __mmask8 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_movdqa64load512_mask ((const __v8di *) __P,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_load_epi64 (__mmask8 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_movdqa64load512_mask ((const __v8di *) __P,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_store_epi64 (void *__P, __m512i __A)
{
  *(__m512i *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_store_epi64 (void *__P, __mmask8 __U, __m512i __A)
{
  __builtin_ia32_movdqa64store512_mask ((__v8di *) __P, (__v8di) __A,
     (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdqa32_512_mask ((__v16si) __A,
           (__v16si) __W,
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdqa32_512_mask ((__v16si) __A,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_load_si512 (void const *__P)
{
  return *(__m512i *) __P;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_load_epi32 (void const *__P)
{
  return *(__m512i *) __P;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_load_epi32 (__m512i __W, __mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_movdqa32load512_mask ((const __v16si *) __P,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_load_epi32 (__mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_movdqa32load512_mask ((const __v16si *) __P,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_store_si512 (void *__P, __m512i __A)
{
  *(__m512i *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_store_epi32 (void *__P, __m512i __A)
{
  *(__m512i *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_store_epi32 (void *__P, __mmask16 __U, __m512i __A)
{
  __builtin_ia32_movdqa32store512_mask ((__v16si *) __P, (__v16si) __A,
     (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mullo_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A * (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mullo_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulld512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mullo_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulld512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sllv_epi32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sllv_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sllv_epi32 (__mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srav_epi32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srav_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srav_epi32 (__mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srlv_epi32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srlv_epi32 (__m512i __W, __mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srlv_epi32 (__mmask16 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv16si_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v8du) __A + (__v8du) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v8du) __A - (__v8du) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sllv_epi64 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_undefined_pd (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sllv_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sllv_epi64 (__mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psllv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srav_epi64 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srav_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srav_epi64 (__mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrav8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srlv_epi64 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srlv_epi64 (__m512i __W, __mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srlv_epi64 (__mmask8 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_psrlv8di_mask ((__v8di) __X,
       (__v8di) __Y,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A + (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mul_epi32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mul_epi32 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mul_epi32 (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
        (__v16si) __Y,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A - (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mul_epu32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
         (__v16si) __Y,
         (__v8di)
         _mm512_undefined_epi32 (),
         (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mul_epu32 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
         (__v16si) __Y,
         (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mul_epu32 (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
         (__v16si) __Y,
         (__v8di)
         _mm512_setzero_si512 (),
         __M);
}
# 981 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sll_epi64 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sll_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sll_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psllq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}
# 1060 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srl_epi64 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srl_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srl_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}
# 1139 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sra_epi64 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sra_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sra_epi64 (__mmask8 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psraq512_mask ((__v8di) __A,
       (__v2di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}
# 1218 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sll_epi32 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_undefined_epi32 (),
       (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sll_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sll_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_pslld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}
# 1297 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srl_epi32 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_undefined_epi32 (),
       (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srl_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srl_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrld512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}
# 1376 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sra_epi32 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_undefined_epi32 (),
       (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sra_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sra_epi32 (__mmask16 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrad512_mask ((__v16si) __A,
       (__v4si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}
# 1549 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rcp14_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
         (__v8df)
         _mm512_undefined_pd (),
         (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rcp14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
         (__v8df) __W,
         (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rcp14_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rcp14pd512_mask ((__v8df) __A,
         (__v8df)
         _mm512_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rcp14_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
        (__v16sf)
        _mm512_undefined_ps (),
        (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rcp14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
        (__v16sf) __W,
        (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rcp14_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rcp14ps512_mask ((__v16sf) __A,
        (__v16sf)
        _mm512_setzero_ps (),
        (__mmask16) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp14_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_rcp14sd ((__v2df) __B,
        (__v2df) __A);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp14_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_rcp14ss ((__v4sf) __B,
       (__v4sf) __A);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rsqrt14_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
           (__v8df)
           _mm512_undefined_pd (),
           (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rsqrt14_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
           (__v8df) __W,
           (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rsqrt14_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rsqrt14pd512_mask ((__v8df) __A,
           (__v8df)
           _mm512_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rsqrt14_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
          (__v16sf)
          _mm512_undefined_ps (),
          (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rsqrt14_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
          (__v16sf) __W,
          (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rsqrt14_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rsqrt14ps512_mask ((__v16sf) __A,
          (__v16sf)
          _mm512_setzero_ps (),
          (__mmask16) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt14_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_rsqrt14sd ((__v2df) __B,
          (__v2df) __A);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt14_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_rsqrt14ss ((__v4sf) __B,
         (__v4sf) __A);
}
# 1800 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi8_epi32 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbd512_mask ((__v16qi) __A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi8_epi32 (__m512i __W, __mmask16 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbd512_mask ((__v16qi) __A,
          (__v16si) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi8_epi32 (__mmask16 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbd512_mask ((__v16qi) __A,
          (__v16si)
          _mm512_setzero_si512 (),
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi8_epi64 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbq512_mask ((__v16qi) __A,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi8_epi64 (__m512i __W, __mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbq512_mask ((__v16qi) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbq512_mask ((__v16qi) __A,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi16_epi32 (__m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwd512_mask ((__v16hi) __A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi16_epi32 (__m512i __W, __mmask16 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwd512_mask ((__v16hi) __A,
          (__v16si) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi16_epi32 (__mmask16 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwd512_mask ((__v16hi) __A,
          (__v16si)
          _mm512_setzero_si512 (),
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi16_epi64 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwq512_mask ((__v8hi) __A,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi16_epi64 (__m512i __W, __mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwq512_mask ((__v8hi) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovsxwq512_mask ((__v8hi) __A,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi32_epi64 (__m256i __X)
{
  return (__m512i) __builtin_ia32_pmovsxdq512_mask ((__v8si) __X,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_epi64 (__m512i __W, __mmask8 __U, __m256i __X)
{
  return (__m512i) __builtin_ia32_pmovsxdq512_mask ((__v8si) __X,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi32_epi64 (__mmask8 __U, __m256i __X)
{
  return (__m512i) __builtin_ia32_pmovsxdq512_mask ((__v8si) __X,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu8_epi32 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbd512_mask ((__v16qi) __A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu8_epi32 (__m512i __W, __mmask16 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbd512_mask ((__v16qi) __A,
          (__v16si) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu8_epi32 (__mmask16 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbd512_mask ((__v16qi) __A,
          (__v16si)
          _mm512_setzero_si512 (),
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu8_epi64 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbq512_mask ((__v16qi) __A,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu8_epi64 (__m512i __W, __mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbq512_mask ((__v16qi) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbq512_mask ((__v16qi) __A,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu16_epi32 (__m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwd512_mask ((__v16hi) __A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu16_epi32 (__m512i __W, __mmask16 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwd512_mask ((__v16hi) __A,
          (__v16si) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu16_epi32 (__mmask16 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwd512_mask ((__v16hi) __A,
          (__v16si)
          _mm512_setzero_si512 (),
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu16_epi64 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwq512_mask ((__v8hi) __A,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu16_epi64 (__m512i __W, __mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwq512_mask ((__v8hi) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m512i) __builtin_ia32_pmovzxwq512_mask ((__v8hi) __A,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu32_epi64 (__m256i __X)
{
  return (__m512i) __builtin_ia32_pmovzxdq512_mask ((__v8si) __X,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu32_epi64 (__m512i __W, __mmask8 __U, __m256i __X)
{
  return (__m512i) __builtin_ia32_pmovzxdq512_mask ((__v8si) __X,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu32_epi64 (__mmask8 __U, __m256i __X)
{
  return (__m512i) __builtin_ia32_pmovzxdq512_mask ((__v8si) __X,
          (__v8di)
          _mm512_setzero_si512 (),
          (__mmask8) __U);
}
# 3429 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_epi64 (__m512i __A)
{
  return (__m512i) __builtin_ia32_pabsq512_mask ((__v8di) __A,
       (__v8di)
       _mm512_undefined_epi32 (),
       (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsq512_mask ((__v8di) __A,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_abs_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsq512_mask ((__v8di) __A,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_epi32 (__m512i __A)
{
  return (__m512i) __builtin_ia32_pabsd512_mask ((__v16si) __A,
       (__v16si)
       _mm512_undefined_epi32 (),
       (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsd512_mask ((__v16si) __A,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_abs_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsd512_mask ((__v16si) __A,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastss_ps (__m128 __A)
{
  return (__m512) __builtin_ia32_broadcastss512 ((__v4sf) __A,
       (__v16sf)
       _mm512_undefined_ps (),
       (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastss_ps (__m512 __O, __mmask16 __M, __m128 __A)
{
  return (__m512) __builtin_ia32_broadcastss512 ((__v4sf) __A,
       (__v16sf) __O, __M);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastss_ps (__mmask16 __M, __m128 __A)
{
  return (__m512) __builtin_ia32_broadcastss512 ((__v4sf) __A,
       (__v16sf)
       _mm512_setzero_ps (),
       __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastsd_pd (__m128d __A)
{
  return (__m512d) __builtin_ia32_broadcastsd512 ((__v2df) __A,
        (__v8df)
        _mm512_undefined_pd (),
        (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastsd_pd (__m512d __O, __mmask8 __M, __m128d __A)
{
  return (__m512d) __builtin_ia32_broadcastsd512 ((__v2df) __A,
        (__v8df) __O, __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastsd_pd (__mmask8 __M, __m128d __A)
{
  return (__m512d) __builtin_ia32_broadcastsd512 ((__v2df) __A,
        (__v8df)
        _mm512_setzero_pd (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastd_epi32 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastd512 ((__v4si) __A,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastd_epi32 (__m512i __O, __mmask16 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastd512 ((__v4si) __A,
        (__v16si) __O, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastd_epi32 (__mmask16 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastd512 ((__v4si) __A,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_epi32 (int __A)
{
  return (__m512i) __builtin_ia32_pbroadcastd512_gpr_mask (__A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16)(-1));
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_set1_epi32 (__m512i __O, __mmask16 __M, int __A)
{
  return (__m512i) __builtin_ia32_pbroadcastd512_gpr_mask (__A, (__v16si) __O,
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_set1_epi32 (__mmask16 __M, int __A)
{
  return (__m512i)
  __builtin_ia32_pbroadcastd512_gpr_mask (__A,
       (__v16si) _mm512_setzero_si512 (),
       __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastq_epi64 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastq512 ((__v2di) __A,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastq_epi64 (__m512i __O, __mmask8 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastq512 ((__v2di) __A,
        (__v8di) __O, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastq_epi64 (__mmask8 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastq512 ((__v2di) __A,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_set1_epi64 (long long __A)
{
  return (__m512i) __builtin_ia32_pbroadcastq512_gpr_mask (__A,
          (__v8di)
          _mm512_undefined_epi32 (),
          (__mmask8)(-1));
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_set1_epi64 (__m512i __O, __mmask8 __M, long long __A)
{
  return (__m512i) __builtin_ia32_pbroadcastq512_gpr_mask (__A, (__v8di) __O,
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_set1_epi64 (__mmask8 __M, long long __A)
{
  return (__m512i)
  __builtin_ia32_pbroadcastq512_gpr_mask (__A,
       (__v8di) _mm512_setzero_si512 (),
       __M);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcast_f32x4 (__m128 __A)
{
  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
           (__v16sf)
           _mm512_undefined_ps (),
           (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcast_f32x4 (__m512 __O, __mmask16 __M, __m128 __A)
{
  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
           (__v16sf) __O,
           __M);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcast_f32x4 (__mmask16 __M, __m128 __A)
{
  return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
           (__v16sf)
           _mm512_setzero_ps (),
           __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcast_i32x4 (__m128i __A)
{
  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
            (__v16si)
            _mm512_undefined_epi32 (),
            (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcast_i32x4 (__m512i __O, __mmask16 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
            (__v16si) __O,
            __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcast_i32x4 (__mmask16 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
            (__v16si)
            _mm512_setzero_si512 (),
            __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcast_f64x4 (__m256d __A)
{
  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
            (__v8df)
            _mm512_undefined_pd (),
            (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcast_f64x4 (__m512d __O, __mmask8 __M, __m256d __A)
{
  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
            (__v8df) __O,
            __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcast_f64x4 (__mmask8 __M, __m256d __A)
{
  return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
            (__v8df)
            _mm512_setzero_pd (),
            __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcast_i64x4 (__m256i __A)
{
  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
            (__v8di)
            _mm512_undefined_epi32 (),
            (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcast_i64x4 (__m512i __O, __mmask8 __M, __m256i __A)
{
  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
            (__v8di) __O,
            __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcast_i64x4 (__mmask8 __M, __m256i __A)
{
  return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
            (__v8di)
            _mm512_setzero_si512 (),
            __M);
}

typedef enum
{
  _MM_PERM_AAAA = 0x00, _MM_PERM_AAAB = 0x01, _MM_PERM_AAAC = 0x02,
  _MM_PERM_AAAD = 0x03, _MM_PERM_AABA = 0x04, _MM_PERM_AABB = 0x05,
  _MM_PERM_AABC = 0x06, _MM_PERM_AABD = 0x07, _MM_PERM_AACA = 0x08,
  _MM_PERM_AACB = 0x09, _MM_PERM_AACC = 0x0A, _MM_PERM_AACD = 0x0B,
  _MM_PERM_AADA = 0x0C, _MM_PERM_AADB = 0x0D, _MM_PERM_AADC = 0x0E,
  _MM_PERM_AADD = 0x0F, _MM_PERM_ABAA = 0x10, _MM_PERM_ABAB = 0x11,
  _MM_PERM_ABAC = 0x12, _MM_PERM_ABAD = 0x13, _MM_PERM_ABBA = 0x14,
  _MM_PERM_ABBB = 0x15, _MM_PERM_ABBC = 0x16, _MM_PERM_ABBD = 0x17,
  _MM_PERM_ABCA = 0x18, _MM_PERM_ABCB = 0x19, _MM_PERM_ABCC = 0x1A,
  _MM_PERM_ABCD = 0x1B, _MM_PERM_ABDA = 0x1C, _MM_PERM_ABDB = 0x1D,
  _MM_PERM_ABDC = 0x1E, _MM_PERM_ABDD = 0x1F, _MM_PERM_ACAA = 0x20,
  _MM_PERM_ACAB = 0x21, _MM_PERM_ACAC = 0x22, _MM_PERM_ACAD = 0x23,
  _MM_PERM_ACBA = 0x24, _MM_PERM_ACBB = 0x25, _MM_PERM_ACBC = 0x26,
  _MM_PERM_ACBD = 0x27, _MM_PERM_ACCA = 0x28, _MM_PERM_ACCB = 0x29,
  _MM_PERM_ACCC = 0x2A, _MM_PERM_ACCD = 0x2B, _MM_PERM_ACDA = 0x2C,
  _MM_PERM_ACDB = 0x2D, _MM_PERM_ACDC = 0x2E, _MM_PERM_ACDD = 0x2F,
  _MM_PERM_ADAA = 0x30, _MM_PERM_ADAB = 0x31, _MM_PERM_ADAC = 0x32,
  _MM_PERM_ADAD = 0x33, _MM_PERM_ADBA = 0x34, _MM_PERM_ADBB = 0x35,
  _MM_PERM_ADBC = 0x36, _MM_PERM_ADBD = 0x37, _MM_PERM_ADCA = 0x38,
  _MM_PERM_ADCB = 0x39, _MM_PERM_ADCC = 0x3A, _MM_PERM_ADCD = 0x3B,
  _MM_PERM_ADDA = 0x3C, _MM_PERM_ADDB = 0x3D, _MM_PERM_ADDC = 0x3E,
  _MM_PERM_ADDD = 0x3F, _MM_PERM_BAAA = 0x40, _MM_PERM_BAAB = 0x41,
  _MM_PERM_BAAC = 0x42, _MM_PERM_BAAD = 0x43, _MM_PERM_BABA = 0x44,
  _MM_PERM_BABB = 0x45, _MM_PERM_BABC = 0x46, _MM_PERM_BABD = 0x47,
  _MM_PERM_BACA = 0x48, _MM_PERM_BACB = 0x49, _MM_PERM_BACC = 0x4A,
  _MM_PERM_BACD = 0x4B, _MM_PERM_BADA = 0x4C, _MM_PERM_BADB = 0x4D,
  _MM_PERM_BADC = 0x4E, _MM_PERM_BADD = 0x4F, _MM_PERM_BBAA = 0x50,
  _MM_PERM_BBAB = 0x51, _MM_PERM_BBAC = 0x52, _MM_PERM_BBAD = 0x53,
  _MM_PERM_BBBA = 0x54, _MM_PERM_BBBB = 0x55, _MM_PERM_BBBC = 0x56,
  _MM_PERM_BBBD = 0x57, _MM_PERM_BBCA = 0x58, _MM_PERM_BBCB = 0x59,
  _MM_PERM_BBCC = 0x5A, _MM_PERM_BBCD = 0x5B, _MM_PERM_BBDA = 0x5C,
  _MM_PERM_BBDB = 0x5D, _MM_PERM_BBDC = 0x5E, _MM_PERM_BBDD = 0x5F,
  _MM_PERM_BCAA = 0x60, _MM_PERM_BCAB = 0x61, _MM_PERM_BCAC = 0x62,
  _MM_PERM_BCAD = 0x63, _MM_PERM_BCBA = 0x64, _MM_PERM_BCBB = 0x65,
  _MM_PERM_BCBC = 0x66, _MM_PERM_BCBD = 0x67, _MM_PERM_BCCA = 0x68,
  _MM_PERM_BCCB = 0x69, _MM_PERM_BCCC = 0x6A, _MM_PERM_BCCD = 0x6B,
  _MM_PERM_BCDA = 0x6C, _MM_PERM_BCDB = 0x6D, _MM_PERM_BCDC = 0x6E,
  _MM_PERM_BCDD = 0x6F, _MM_PERM_BDAA = 0x70, _MM_PERM_BDAB = 0x71,
  _MM_PERM_BDAC = 0x72, _MM_PERM_BDAD = 0x73, _MM_PERM_BDBA = 0x74,
  _MM_PERM_BDBB = 0x75, _MM_PERM_BDBC = 0x76, _MM_PERM_BDBD = 0x77,
  _MM_PERM_BDCA = 0x78, _MM_PERM_BDCB = 0x79, _MM_PERM_BDCC = 0x7A,
  _MM_PERM_BDCD = 0x7B, _MM_PERM_BDDA = 0x7C, _MM_PERM_BDDB = 0x7D,
  _MM_PERM_BDDC = 0x7E, _MM_PERM_BDDD = 0x7F, _MM_PERM_CAAA = 0x80,
  _MM_PERM_CAAB = 0x81, _MM_PERM_CAAC = 0x82, _MM_PERM_CAAD = 0x83,
  _MM_PERM_CABA = 0x84, _MM_PERM_CABB = 0x85, _MM_PERM_CABC = 0x86,
  _MM_PERM_CABD = 0x87, _MM_PERM_CACA = 0x88, _MM_PERM_CACB = 0x89,
  _MM_PERM_CACC = 0x8A, _MM_PERM_CACD = 0x8B, _MM_PERM_CADA = 0x8C,
  _MM_PERM_CADB = 0x8D, _MM_PERM_CADC = 0x8E, _MM_PERM_CADD = 0x8F,
  _MM_PERM_CBAA = 0x90, _MM_PERM_CBAB = 0x91, _MM_PERM_CBAC = 0x92,
  _MM_PERM_CBAD = 0x93, _MM_PERM_CBBA = 0x94, _MM_PERM_CBBB = 0x95,
  _MM_PERM_CBBC = 0x96, _MM_PERM_CBBD = 0x97, _MM_PERM_CBCA = 0x98,
  _MM_PERM_CBCB = 0x99, _MM_PERM_CBCC = 0x9A, _MM_PERM_CBCD = 0x9B,
  _MM_PERM_CBDA = 0x9C, _MM_PERM_CBDB = 0x9D, _MM_PERM_CBDC = 0x9E,
  _MM_PERM_CBDD = 0x9F, _MM_PERM_CCAA = 0xA0, _MM_PERM_CCAB = 0xA1,
  _MM_PERM_CCAC = 0xA2, _MM_PERM_CCAD = 0xA3, _MM_PERM_CCBA = 0xA4,
  _MM_PERM_CCBB = 0xA5, _MM_PERM_CCBC = 0xA6, _MM_PERM_CCBD = 0xA7,
  _MM_PERM_CCCA = 0xA8, _MM_PERM_CCCB = 0xA9, _MM_PERM_CCCC = 0xAA,
  _MM_PERM_CCCD = 0xAB, _MM_PERM_CCDA = 0xAC, _MM_PERM_CCDB = 0xAD,
  _MM_PERM_CCDC = 0xAE, _MM_PERM_CCDD = 0xAF, _MM_PERM_CDAA = 0xB0,
  _MM_PERM_CDAB = 0xB1, _MM_PERM_CDAC = 0xB2, _MM_PERM_CDAD = 0xB3,
  _MM_PERM_CDBA = 0xB4, _MM_PERM_CDBB = 0xB5, _MM_PERM_CDBC = 0xB6,
  _MM_PERM_CDBD = 0xB7, _MM_PERM_CDCA = 0xB8, _MM_PERM_CDCB = 0xB9,
  _MM_PERM_CDCC = 0xBA, _MM_PERM_CDCD = 0xBB, _MM_PERM_CDDA = 0xBC,
  _MM_PERM_CDDB = 0xBD, _MM_PERM_CDDC = 0xBE, _MM_PERM_CDDD = 0xBF,
  _MM_PERM_DAAA = 0xC0, _MM_PERM_DAAB = 0xC1, _MM_PERM_DAAC = 0xC2,
  _MM_PERM_DAAD = 0xC3, _MM_PERM_DABA = 0xC4, _MM_PERM_DABB = 0xC5,
  _MM_PERM_DABC = 0xC6, _MM_PERM_DABD = 0xC7, _MM_PERM_DACA = 0xC8,
  _MM_PERM_DACB = 0xC9, _MM_PERM_DACC = 0xCA, _MM_PERM_DACD = 0xCB,
  _MM_PERM_DADA = 0xCC, _MM_PERM_DADB = 0xCD, _MM_PERM_DADC = 0xCE,
  _MM_PERM_DADD = 0xCF, _MM_PERM_DBAA = 0xD0, _MM_PERM_DBAB = 0xD1,
  _MM_PERM_DBAC = 0xD2, _MM_PERM_DBAD = 0xD3, _MM_PERM_DBBA = 0xD4,
  _MM_PERM_DBBB = 0xD5, _MM_PERM_DBBC = 0xD6, _MM_PERM_DBBD = 0xD7,
  _MM_PERM_DBCA = 0xD8, _MM_PERM_DBCB = 0xD9, _MM_PERM_DBCC = 0xDA,
  _MM_PERM_DBCD = 0xDB, _MM_PERM_DBDA = 0xDC, _MM_PERM_DBDB = 0xDD,
  _MM_PERM_DBDC = 0xDE, _MM_PERM_DBDD = 0xDF, _MM_PERM_DCAA = 0xE0,
  _MM_PERM_DCAB = 0xE1, _MM_PERM_DCAC = 0xE2, _MM_PERM_DCAD = 0xE3,
  _MM_PERM_DCBA = 0xE4, _MM_PERM_DCBB = 0xE5, _MM_PERM_DCBC = 0xE6,
  _MM_PERM_DCBD = 0xE7, _MM_PERM_DCCA = 0xE8, _MM_PERM_DCCB = 0xE9,
  _MM_PERM_DCCC = 0xEA, _MM_PERM_DCCD = 0xEB, _MM_PERM_DCDA = 0xEC,
  _MM_PERM_DCDB = 0xED, _MM_PERM_DCDC = 0xEE, _MM_PERM_DCDD = 0xEF,
  _MM_PERM_DDAA = 0xF0, _MM_PERM_DDAB = 0xF1, _MM_PERM_DDAC = 0xF2,
  _MM_PERM_DDAD = 0xF3, _MM_PERM_DDBA = 0xF4, _MM_PERM_DDBB = 0xF5,
  _MM_PERM_DDBC = 0xF6, _MM_PERM_DDBD = 0xF7, _MM_PERM_DDCA = 0xF8,
  _MM_PERM_DDCB = 0xF9, _MM_PERM_DDCC = 0xFA, _MM_PERM_DDCD = 0xFB,
  _MM_PERM_DDDA = 0xFC, _MM_PERM_DDDB = 0xFD, _MM_PERM_DDDC = 0xFE,
  _MM_PERM_DDDD = 0xFF
} _MM_PERM_ENUM;
# 4123 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rolv_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rolv_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rolv_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rorv_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rorv_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rorv_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rolv_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rolv_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W,
        (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rolv_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prolvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_rorv_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_rorv_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W,
        (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_rorv_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_prorvq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        (__mmask8) __U);
}
# 4571 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtu32_sd (__m128d __A, unsigned __B)
{
  return (__m128d) __builtin_ia32_cvtusi2sd32 ((__v2df) __A, __B);
}
# 4680 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi32_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_storeu_epi8 (void * __P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovdb512mem_mask ((__v16qi *) __P, (__v16si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
        (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi32_epi8 (__mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi32_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb512_mask ((__v16si) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi32_storeu_epi8 (void * __P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovsdb512mem_mask ((__v16qi *) __P, (__v16si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb512_mask ((__v16si) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi32_epi8 (__mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb512_mask ((__v16si) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi32_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb512_mask ((__v16si) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi32_storeu_epi8 (void * __P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovusdb512mem_mask ((__v16qi *) __P, (__v16si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb512_mask ((__v16si) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi32_epi8 (__mmask16 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb512_mask ((__v16si) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi32_epi16 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
        (__v16hi)
        _mm256_undefined_si256 (),
        (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_storeu_epi16 (void * __P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovdw512mem_mask ((__v16hi *) __P, (__v16si) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
        (__v16hi) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi32_epi16 (__mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
        (__v16hi)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi32_epi16 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsdw512_mask ((__v16si) __A,
         (__v16hi)
         _mm256_undefined_si256 (),
         (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi32_storeu_epi16 (void *__P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovsdw512mem_mask ((__v16hi*) __P, (__v16si) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsdw512_mask ((__v16si) __A,
         (__v16hi) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi32_epi16 (__mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsdw512_mask ((__v16si) __A,
         (__v16hi)
         _mm256_setzero_si256 (),
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi32_epi16 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusdw512_mask ((__v16si) __A,
          (__v16hi)
          _mm256_undefined_si256 (),
          (__mmask16) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi32_storeu_epi16 (void *__P, __mmask16 __M, __m512i __A)
{
  __builtin_ia32_pmovusdw512mem_mask ((__v16hi*) __P, (__v16si) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusdw512_mask ((__v16si) __A,
          (__v16hi) __O,
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi32_epi16 (__mmask16 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusdw512_mask ((__v16si) __A,
          (__v16hi)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi64_epi32 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
        (__v8si)
        _mm256_undefined_si256 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_storeu_epi32 (void* __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovqd512mem_mask ((__v8si *) __P, (__v8di) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_epi32 (__m256i __O, __mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
        (__v8si) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi64_epi32 (__mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovqd512_mask ((__v8di) __A,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi64_epi32 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsqd512_mask ((__v8di) __A,
         (__v8si)
         _mm256_undefined_si256 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_storeu_epi32 (void *__P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovsqd512mem_mask ((__v8si *) __P, (__v8di) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_epi32 (__m256i __O, __mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsqd512_mask ((__v8di) __A,
         (__v8si) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi64_epi32 (__mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovsqd512_mask ((__v8di) __A,
         (__v8si)
         _mm256_setzero_si256 (),
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi64_epi32 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusqd512_mask ((__v8di) __A,
          (__v8si)
          _mm256_undefined_si256 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_storeu_epi32 (void* __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovusqd512mem_mask ((__v8si*) __P, (__v8di) __A, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_epi32 (__m256i __O, __mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusqd512_mask ((__v8di) __A,
          (__v8si) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi64_epi32 (__mmask8 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovusqd512_mask ((__v8di) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi64_epi16 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
        (__v8hi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovqw512mem_mask ((__v8hi *) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
        (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi64_epi16 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi64_epi16 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw512_mask ((__v8di) __A,
         (__v8hi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_storeu_epi16 (void * __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovsqw512mem_mask ((__v8hi *) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw512_mask ((__v8di) __A,
         (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi64_epi16 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw512_mask ((__v8di) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi64_epi16 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw512_mask ((__v8di) __A,
          (__v8hi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovusqw512mem_mask ((__v8hi*) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw512_mask ((__v8di) __A,
          (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi64_epi16 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw512_mask ((__v8di) __A,
          (__v8hi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi64_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovqb512mem_mask ((__v16qi *) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A,
        (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi64_epi8 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovqb512_mask ((__v8di) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi64_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb512_mask ((__v8di) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_storeu_epi8 (void * __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovsqb512mem_mask ((__v16qi *) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi64_epi8 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb512_mask ((__v8di) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi64_epi8 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb512_mask ((__v8di) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi64_epi8 (__m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb512_mask ((__v8di) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_storeu_epi8 (void * __P, __mmask8 __M, __m512i __A)
{
  __builtin_ia32_pmovusqb512mem_mask ((__v16qi *) __P, (__v8di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi64_epi8 (__m128i __O, __mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb512_mask ((__v8di) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi64_epi8 (__mmask8 __M, __m512i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb512_mask ((__v8di) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi32_pd (__m256i __A)
{
  return (__m512d) __builtin_ia32_cvtdq2pd512_mask ((__v8si) __A,
          (__v8df)
          _mm512_undefined_pd (),
          (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_pd (__m512d __W, __mmask8 __U, __m256i __A)
{
  return (__m512d) __builtin_ia32_cvtdq2pd512_mask ((__v8si) __A,
          (__v8df) __W,
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi32_pd (__mmask8 __U, __m256i __A)
{
  return (__m512d) __builtin_ia32_cvtdq2pd512_mask ((__v8si) __A,
          (__v8df)
          _mm512_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu32_pd (__m256i __A)
{
  return (__m512d) __builtin_ia32_cvtudq2pd512_mask ((__v8si) __A,
           (__v8df)
           _mm512_undefined_pd (),
           (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu32_pd (__m512d __W, __mmask8 __U, __m256i __A)
{
  return (__m512d) __builtin_ia32_cvtudq2pd512_mask ((__v8si) __A,
           (__v8df) __W,
           (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu32_pd (__mmask8 __U, __m256i __A)
{
  return (__m512d) __builtin_ia32_cvtudq2pd512_mask ((__v8si) __A,
           (__v8df)
           _mm512_setzero_pd (),
           (__mmask8) __U);
}
# 5694 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_loadu_pd (void const *__P)
{
  return *(__m512d_u *)__P;
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_pd (__m512d __W, __mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_loadupd512_mask ((const double *) __P,
         (__v8df) __W,
         (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_pd (__mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_loadupd512_mask ((const double *) __P,
         (__v8df)
         _mm512_setzero_pd (),
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_storeu_pd (void *__P, __m512d __A)
{
  *(__m512d_u *)__P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_pd (void *__P, __mmask8 __U, __m512d __A)
{
  __builtin_ia32_storeupd512_mask ((double *) __P, (__v8df) __A,
       (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_loadu_ps (void const *__P)
{
  return *(__m512_u *)__P;
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_ps (__m512 __W, __mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_loadups512_mask ((const float *) __P,
        (__v16sf) __W,
        (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_ps (__mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_loadups512_mask ((const float *) __P,
        (__v16sf)
        _mm512_setzero_ps (),
        (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_storeu_ps (void *__P, __m512 __A)
{
  *(__m512_u *)__P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_ps (void *__P, __mmask16 __U, __m512 __A)
{
  __builtin_ia32_storeups512_mask ((float *) __P, (__v16sf) __A,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_epi64 (__m512i __W, __mmask8 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddqudi512_mask ((const long long *) __P,
           (__v8di) __W,
           (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddqudi512_mask ((const long long *) __P,
           (__v8di)
           _mm512_setzero_si512 (),
           (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_epi64 (void *__P, __mmask8 __U, __m512i __A)
{
  __builtin_ia32_storedqudi512_mask ((long long *) __P, (__v8di) __A,
         (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_loadu_si512 (void const *__P)
{
  return *(__m512i_u *)__P;
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_epi32 (__m512i __W, __mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddqusi512_mask ((const int *) __P,
           (__v16si) __W,
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_epi32 (__mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddqusi512_mask ((const int *) __P,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_storeu_si512 (void *__P, __m512i __A)
{
  *(__m512i_u *)__P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_epi32 (void *__P, __mmask16 __U, __m512i __A)
{
  __builtin_ia32_storedqusi512_mask ((int *) __P, (__v16si) __A,
         (__mmask16) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutevar_pd (__m512d __A, __m512i __C)
{
  return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
       (__v8di) __C,
       (__v8df)
       _mm512_undefined_pd (),
       (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutevar_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512i __C)
{
  return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
       (__v8di) __C,
       (__v8df) __W,
       (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutevar_pd (__mmask8 __U, __m512d __A, __m512i __C)
{
  return (__m512d) __builtin_ia32_vpermilvarpd512_mask ((__v8df) __A,
       (__v8di) __C,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutevar_ps (__m512 __A, __m512i __C)
{
  return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
             (__v16si) __C,
             (__v16sf)
             _mm512_undefined_ps (),
             (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutevar_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512i __C)
{
  return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
             (__v16si) __C,
             (__v16sf) __W,
             (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutevar_ps (__mmask16 __U, __m512 __A, __m512i __C)
{
  return (__m512) __builtin_ia32_vpermilvarps512_mask ((__v16sf) __A,
             (__v16si) __C,
             (__v16sf)
             _mm512_setzero_ps (),
             (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutex2var_epi64 (__m512i __A, __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I
                       ,
             (__v8di) __A,
             (__v8di) __B,
             (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutex2var_epi64 (__m512i __A, __mmask8 __U, __m512i __I,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varq512_mask ((__v8di) __I
                       ,
             (__v8di) __A,
             (__v8di) __B,
             (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2_permutex2var_epi64 (__m512i __A, __m512i __I,
     __mmask8 __U, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermi2varq512_mask ((__v8di) __A,
             (__v8di) __I
                       ,
             (__v8di) __B,
             (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutex2var_epi64 (__mmask8 __U, __m512i __A,
     __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varq512_maskz ((__v8di) __I
                 ,
       (__v8di) __A,
       (__v8di) __B,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutex2var_epi32 (__m512i __A, __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I
                       ,
             (__v16si) __A,
             (__v16si) __B,
             (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutex2var_epi32 (__m512i __A, __mmask16 __U,
    __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2vard512_mask ((__v16si) __I
                       ,
             (__v16si) __A,
             (__v16si) __B,
             (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2_permutex2var_epi32 (__m512i __A, __m512i __I,
     __mmask16 __U, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermi2vard512_mask ((__v16si) __A,
             (__v16si) __I
                       ,
             (__v16si) __B,
             (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutex2var_epi32 (__mmask16 __U, __m512i __A,
     __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2vard512_maskz ((__v16si) __I
                 ,
       (__v16si) __A,
       (__v16si) __B,
       (__mmask16) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutex2var_pd (__m512d __A, __m512i __I, __m512d __B)
{
  return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I
                 ,
       (__v8df) __A,
       (__v8df) __B,
       (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutex2var_pd (__m512d __A, __mmask8 __U, __m512i __I,
        __m512d __B)
{
  return (__m512d) __builtin_ia32_vpermt2varpd512_mask ((__v8di) __I
                 ,
       (__v8df) __A,
       (__v8df) __B,
       (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2_permutex2var_pd (__m512d __A, __m512i __I, __mmask8 __U,
         __m512d __B)
{
  return (__m512d) __builtin_ia32_vpermi2varpd512_mask ((__v8df) __A,
       (__v8di) __I
                 ,
       (__v8df) __B,
       (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutex2var_pd (__mmask8 __U, __m512d __A, __m512i __I,
         __m512d __B)
{
  return (__m512d) __builtin_ia32_vpermt2varpd512_maskz ((__v8di) __I
                  ,
        (__v8df) __A,
        (__v8df) __B,
        (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutex2var_ps (__m512 __A, __m512i __I, __m512 __B)
{
  return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I
                       ,
             (__v16sf) __A,
             (__v16sf) __B,
             (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutex2var_ps (__m512 __A, __mmask16 __U, __m512i __I, __m512 __B)
{
  return (__m512) __builtin_ia32_vpermt2varps512_mask ((__v16si) __I
                       ,
             (__v16sf) __A,
             (__v16sf) __B,
             (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2_permutex2var_ps (__m512 __A, __m512i __I, __mmask16 __U,
         __m512 __B)
{
  return (__m512) __builtin_ia32_vpermi2varps512_mask ((__v16sf) __A,
             (__v16si) __I
                       ,
             (__v16sf) __B,
             (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutex2var_ps (__mmask16 __U, __m512 __A, __m512i __I,
         __m512 __B)
{
  return (__m512) __builtin_ia32_vpermt2varps512_maskz ((__v16si) __I
                 ,
       (__v16sf) __A,
       (__v16sf) __B,
       (__mmask16) __U);
}
# 6280 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutexvar_epi64 (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvardi512_mask ((__v8di) __Y,
           (__v8di) __X,
           (__v8di)
           _mm512_setzero_si512 (),
           __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutexvar_epi64 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvardi512_mask ((__v8di) __Y,
           (__v8di) __X,
           (__v8di)
           _mm512_undefined_epi32 (),
           (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutexvar_epi64 (__m512i __W, __mmask8 __M, __m512i __X,
          __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvardi512_mask ((__v8di) __Y,
           (__v8di) __X,
           (__v8di) __W,
           __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutexvar_epi32 (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y,
           (__v16si) __X,
           (__v16si)
           _mm512_setzero_si512 (),
           __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutexvar_epi32 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y,
           (__v16si) __X,
           (__v16si)
           _mm512_undefined_epi32 (),
           (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X,
          __m512i __Y)
{
  return (__m512i) __builtin_ia32_permvarsi512_mask ((__v16si) __Y,
           (__v16si) __X,
           (__v16si) __W,
           __M);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutexvar_pd (__m512i __X, __m512d __Y)
{
  return (__m512d) __builtin_ia32_permvardf512_mask ((__v8df) __Y,
           (__v8di) __X,
           (__v8df)
           _mm512_undefined_pd (),
           (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutexvar_pd (__m512d __W, __mmask8 __U, __m512i __X, __m512d __Y)
{
  return (__m512d) __builtin_ia32_permvardf512_mask ((__v8df) __Y,
           (__v8di) __X,
           (__v8df) __W,
           (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutexvar_pd (__mmask8 __U, __m512i __X, __m512d __Y)
{
  return (__m512d) __builtin_ia32_permvardf512_mask ((__v8df) __Y,
           (__v8di) __X,
           (__v8df)
           _mm512_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutexvar_ps (__m512i __X, __m512 __Y)
{
  return (__m512) __builtin_ia32_permvarsf512_mask ((__v16sf) __Y,
          (__v16si) __X,
          (__v16sf)
          _mm512_undefined_ps (),
          (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutexvar_ps (__m512 __W, __mmask16 __U, __m512i __X, __m512 __Y)
{
  return (__m512) __builtin_ia32_permvarsf512_mask ((__v16sf) __Y,
          (__v16si) __X,
          (__v16sf) __W,
          (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutexvar_ps (__mmask16 __U, __m512i __X, __m512 __Y)
{
  return (__m512) __builtin_ia32_permvarsf512_mask ((__v16sf) __Y,
          (__v16si) __X,
          (__v16sf)
          _mm512_setzero_ps (),
          (__mmask16) __U);
}
# 6715 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movehdup_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_movehdup_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
         (__v16sf) __W,
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_movehdup_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_moveldup_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_moveldup_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
         (__v16sf) __W,
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_or_si512 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A | (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_or_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A | (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_or_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pord512_mask ((__v16si) __A,
      (__v16si) __B,
      (__v16si) __W,
      (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_or_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pord512_mask ((__v16si) __A,
      (__v16si) __B,
      (__v16si)
      _mm512_setzero_si512 (),
      (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_or_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v8du) __A | (__v8du) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_or_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_porq512_mask ((__v8di) __A,
      (__v8di) __B,
      (__v8di) __W,
      (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_or_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_porq512_mask ((__v8di) __A,
      (__v8di) __B,
      (__v8di)
      _mm512_setzero_si512 (),
      (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_xor_si512 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A ^ (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_xor_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A ^ (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_xor_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pxord512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_xor_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pxord512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_xor_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v8du) __A ^ (__v8du) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_xor_epi64 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pxorq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_xor_epi64 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pxorq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di)
       _mm512_setzero_si512 (),
       (__mmask8) __U);
}
# 7080 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_and_si512 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A & (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_and_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v16su) __A & (__v16su) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_and_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_and_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandd512_mask ((__v16si) __A,
       (__v16si) __B,
       (__v16si)
       _mm512_setzero_si512 (),
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_and_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v8du) __A & (__v8du) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_and_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di) __W, __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_and_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandq512_mask ((__v8di) __A,
       (__v8di) __B,
       (__v8di)
       _mm512_setzero_pd (),
       __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_andnot_si512 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_andnot_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_andnot_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_andnot_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_andnot_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_andnot_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W, __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_andnot_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_pd (),
        __U);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_test_epi32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A,
      (__v16si) __B,
      (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_test_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ptestmd512 ((__v16si) __A,
      (__v16si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_test_epi64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A,
            (__v8di) __B,
            (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_test_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq512 ((__v8di) __A, (__v8di) __B, __U);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_testn_epi32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ptestnmd512 ((__v16si) __A,
       (__v16si) __B,
       (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_testn_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ptestnmd512 ((__v16si) __A,
       (__v16si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_testn_epi64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq512 ((__v8di) __A,
      (__v8di) __B,
      (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_testn_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq512 ((__v8di) __A,
      (__v8di) __B, __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_ps (__m512 __A)
{
  return (__m512) _mm512_and_epi32 ((__m512i) __A,
        _mm512_set1_epi32 (0x7fffffff));
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) _mm512_mask_and_epi32 ((__m512i) __W, __U, (__m512i) __A,
      _mm512_set1_epi32 (0x7fffffff));
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_pd (__m512 __A)
{
  return (__m512d) _mm512_and_epi64 ((__m512i) __A,
         _mm512_set1_epi64 (0x7fffffffffffffffLL));
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_pd (__m512d __W, __mmask8 __U, __m512 __A)
{
  return (__m512d)
  _mm512_mask_and_epi64 ((__m512i) __W, __U, (__m512i) __A,
    _mm512_set1_epi64 (0x7fffffffffffffffLL));
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhdq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si)
           _mm512_undefined_epi32 (),
           (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_epi32 (__m512i __W, __mmask16 __U, __m512i __A,
       __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhdq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si) __W,
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhdq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di)
            _mm512_undefined_epi32 (),
            (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di) __W,
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di)
            _mm512_setzero_si512 (),
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckldq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si)
           _mm512_undefined_epi32 (),
           (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_epi32 (__m512i __W, __mmask16 __U, __m512i __A,
       __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckldq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si) __W,
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckldq512_mask ((__v16si) __A,
           (__v16si) __B,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di)
            _mm512_undefined_epi32 (),
            (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di) __W,
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklqdq512_mask ((__v8di) __A,
            (__v8di) __B,
            (__v8di)
            _mm512_setzero_si512 (),
            (__mmask8) __U);
}
# 7698 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movedup_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A,
         (__v8df)
         _mm512_undefined_pd (),
         (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_movedup_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A,
         (__v8df) __W,
         (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_movedup_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_movddup512_mask ((__v8df) __A,
         (__v8df)
         _mm512_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_pd (__m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpcklpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_undefined_pd (),
          (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpcklpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df) __W,
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpcklpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_pd (__m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpckhpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_undefined_pd (),
          (__mmask8) -1);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpckhpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df) __W,
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_unpckhpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_ps (__m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpckhps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpckhps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __W,
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpckhps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U);
}
# 8052 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_stream_si512 (__m512i * __P, __m512i __A)
{
  __builtin_ia32_movntdq512 ((__v8di *) __P, (__v8di) __A);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_stream_ps (float *__P, __m512 __A)
{
  __builtin_ia32_movntps512 (__P, (__v16sf) __A);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_stream_pd (double *__P, __m512d __A)
{
  __builtin_ia32_movntpd512 (__P, (__v8df) __A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_stream_load_si512 (void *__P)
{
  return __builtin_ia32_movntdqa512 ((__v8di *)__P);
}


typedef enum
{
  _MM_MANT_NORM_1_2,
  _MM_MANT_NORM_p5_2,
  _MM_MANT_NORM_p5_1,
  _MM_MANT_NORM_p75_1p5
} _MM_MANTISSA_NORM_ENUM;

typedef enum
{
  _MM_MANT_SIGN_src,
  _MM_MANT_SIGN_zero,
  _MM_MANT_SIGN_nan
} _MM_MANTISSA_SIGN_ENUM;
# 8472 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_floor_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
        (0x01 | 0x00),
        (__v16sf) __A, -1,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_floor_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
         (0x01 | 0x00),
         (__v8df) __A, -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_ceil_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
        (0x02 | 0x00),
        (__v16sf) __A, -1,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_ceil_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
         (0x02 | 0x00),
         (__v8df) __A, -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_floor_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
        (0x01 | 0x00),
        (__v16sf) __W, __U,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_floor_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
         (0x01 | 0x00),
         (__v8df) __W, __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_ceil_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A,
        (0x02 | 0x00),
        (__v16sf) __W, __U,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_ceil_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A,
         (0x02 | 0x00),
         (__v8df) __W, __U,
         0x04);
}
# 8651 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epi32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_pcmpeqd512_mask ((__v16si) __A,
           (__v16si) __B,
           (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_pcmpeqd512_mask ((__v16si) __A,
           (__v16si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq512_mask ((__v8di) __A,
          (__v8di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epi64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq512_mask ((__v8di) __A,
          (__v8di) __B,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epi32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_pcmpgtd512_mask ((__v16si) __A,
           (__v16si) __B,
           (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epi32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_pcmpgtd512_mask ((__v16si) __A,
           (__v16si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epi64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq512_mask ((__v8di) __A,
          (__v8di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epi64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq512_mask ((__v8di) __A,
          (__v8di) __B,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epi32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 5,
          (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 5,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 5,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 5,
          (__mmask16) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 5,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epi64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 5,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 5,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 5,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 2,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 2,
          (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 2,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 2,
          (__mmask16) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 2,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 2,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 2,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 2,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 1,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 1,
          (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 1,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 1,
          (__mmask16) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epi64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 1,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 1,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 1,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 1,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epi32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 4,
          (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epi32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_cmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 4,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epu32_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 4,
          (__mmask16) __M);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu32_mask (__m512i __X, __m512i __Y)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __X,
          (__v16si) __Y, 4,
          (__mmask16) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epi64_mask (__mmask16 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 4,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epi64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 4,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epu64_mask (__mmask8 __M, __m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 4,
          (__mmask8) __M);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu64_mask (__m512i __X, __m512i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __X,
          (__v8di) __Y, 4,
          (__mmask8) -1);
}
# 9789 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compress_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_compressdf512_mask ((__v8df) __A,
            (__v8df) __W,
            (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_compress_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_compressdf512_mask ((__v8df) __A,
            (__v8df)
            _mm512_setzero_pd (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compressstoreu_pd (void *__P, __mmask8 __U, __m512d __A)
{
  __builtin_ia32_compressstoredf512_mask ((__v8df *) __P, (__v8df) __A,
       (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compress_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_compresssf512_mask ((__v16sf) __A,
           (__v16sf) __W,
           (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_compress_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_compresssf512_mask ((__v16sf) __A,
           (__v16sf)
           _mm512_setzero_ps (),
           (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compressstoreu_ps (void *__P, __mmask16 __U, __m512 __A)
{
  __builtin_ia32_compressstoresf512_mask ((__v16sf *) __P, (__v16sf) __A,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compress_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_compressdi512_mask ((__v8di) __A,
            (__v8di) __W,
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_compress_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_compressdi512_mask ((__v8di) __A,
            (__v8di)
            _mm512_setzero_si512 (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compressstoreu_epi64 (void *__P, __mmask8 __U, __m512i __A)
{
  __builtin_ia32_compressstoredi512_mask ((__v8di *) __P, (__v8di) __A,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compress_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_compresssi512_mask ((__v16si) __A,
            (__v16si) __W,
            (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_compress_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_compresssi512_mask ((__v16si) __A,
            (__v16si)
            _mm512_setzero_si512 (),
            (__mmask16) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_compressstoreu_epi32 (void *__P, __mmask16 __U, __m512i __A)
{
  __builtin_ia32_compressstoresi512_mask ((__v16si *) __P, (__v16si) __A,
       (__mmask16) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expand_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_expanddf512_mask ((__v8df) __A,
          (__v8df) __W,
          (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expand_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_expanddf512_maskz ((__v8df) __A,
           (__v8df)
           _mm512_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expandloadu_pd (__m512d __W, __mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_expandloaddf512_mask ((const __v8df *) __P,
       (__v8df) __W,
       (__mmask8) __U);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expandloadu_pd (__mmask8 __U, void const *__P)
{
  return (__m512d) __builtin_ia32_expandloaddf512_maskz ((const __v8df *) __P,
        (__v8df)
        _mm512_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expand_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_expandsf512_mask ((__v16sf) __A,
         (__v16sf) __W,
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expand_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_expandsf512_maskz ((__v16sf) __A,
          (__v16sf)
          _mm512_setzero_ps (),
          (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expandloadu_ps (__m512 __W, __mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_expandloadsf512_mask ((const __v16sf *) __P,
             (__v16sf) __W,
             (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expandloadu_ps (__mmask16 __U, void const *__P)
{
  return (__m512) __builtin_ia32_expandloadsf512_maskz ((const __v16sf *) __P,
       (__v16sf)
       _mm512_setzero_ps (),
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expand_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_expanddi512_mask ((__v8di) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expand_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_expanddi512_maskz ((__v8di) __A,
           (__v8di)
           _mm512_setzero_si512 (),
           (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expandloadu_epi64 (__m512i __W, __mmask8 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_expandloaddi512_mask ((const __v8di *) __P,
       (__v8di) __W,
       (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expandloadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m512i)
  __builtin_ia32_expandloaddi512_maskz ((const __v8di *) __P,
            (__v8di)
            _mm512_setzero_si512 (),
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expand_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_expandsi512_mask ((__v16si) __A,
          (__v16si) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expand_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_expandsi512_maskz ((__v16si) __A,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_expandloadu_epi32 (__m512i __W, __mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_expandloadsi512_mask ((const __v16si *) __P,
       (__v16si) __W,
       (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_expandloadu_epi32 (__mmask16 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_expandloadsi512_maskz ((const __v16si *) __P,
        (__v16si)
        _mm512_setzero_si512
        (), (__mmask16) __U);
}
# 10058 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortest_mask16_u8 (__mmask16 __A, __mmask16 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_kortestchi (__A, __B);
  return (unsigned char) __builtin_ia32_kortestzhi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestz_mask16_u8 (__mmask16 __A, __mmask16 __B)
{
  return (unsigned char) __builtin_ia32_kortestzhi ((__mmask16) __A,
          (__mmask16) __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestc_mask16_u8 (__mmask16 __A, __mmask16 __B)
{
  return (unsigned char) __builtin_ia32_kortestchi ((__mmask16) __A,
          (__mmask16) __B);
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtmask16_u32 (__mmask16 __A)
{
  return (unsigned int) __builtin_ia32_kmovw ((__mmask16 ) __A);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtu32_mask16 (unsigned int __A)
{
  return (__mmask16) __builtin_ia32_kmovw ((__mmask16 ) __A);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_load_mask16 (__mmask16 *__A)
{
  return (__mmask16) __builtin_ia32_kmovw (*(__mmask16 *) __A);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_store_mask16 (__mmask16 *__A, __mmask16 __B)
{
  *(__mmask16 *) __A = __builtin_ia32_kmovw (__B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kand (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kandhi ((__mmask16) __A, (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kandn (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kandnhi ((__mmask16) __A,
          (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kor (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_korhi ((__mmask16) __A, (__mmask16) __B);
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kortestz (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kortestzhi ((__mmask16) __A,
      (__mmask16) __B);
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kortestc (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kortestchi ((__mmask16) __A,
      (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kxnor (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kxnorhi ((__mmask16) __A, (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kxor (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kxorhi ((__mmask16) __A, (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_knot (__mmask16 __A)
{
  return (__mmask16) __builtin_ia32_knothi ((__mmask16) __A);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kunpackb (__mmask16 __A, __mmask16 __B)
{
  return (__mmask16) __builtin_ia32_kunpckhi ((__mmask16) __A, (__mmask16) __B);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kunpackb_mask16 (__mmask8 __A, __mmask8 __B)
{
  return (__mmask16) __builtin_ia32_kunpckhi ((__mmask16) __A, (__mmask16) __B);
}
# 10253 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epi64 (__mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epi64 (__m512i __W, __mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epi64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epi64 (__m512i __W, __mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epi64 (__mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epu64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epu64 (__mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epu64 (__m512i __W, __mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epu64 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_undefined_epi32 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epu64 (__m512i __W, __mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epu64 (__mmask8 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuq512_mask ((__v8di) __A,
        (__v8di) __B,
        (__v8di)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsd512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epu32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epu32 (__mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epu32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W, __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epu32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_undefined_epi32 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epu32 (__mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si)
        _mm512_setzero_si512 (),
        __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epu32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminud512_mask ((__v16si) __A,
        (__v16si) __B,
        (__v16si) __W, __M);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_ps (__m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpcklps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpcklps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __W,
         (__mmask16) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_unpcklps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U);
}
# 10584 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_blend_pd (__mmask8 __U, __m512d __A, __m512d __W)
{
  return (__m512d) __builtin_ia32_blendmpd_512_mask ((__v8df) __A,
           (__v8df) __W,
           (__mmask8) __U);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_blend_ps (__mmask16 __U, __m512 __A, __m512 __W)
{
  return (__m512) __builtin_ia32_blendmps_512_mask ((__v16sf) __A,
          (__v16sf) __W,
          (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_blend_epi64 (__mmask8 __U, __m512i __A, __m512i __W)
{
  return (__m512i) __builtin_ia32_blendmq_512_mask ((__v8di) __A,
          (__v8di) __W,
          (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_blend_epi32 (__mmask16 __U, __m512i __A, __m512i __W)
{
  return (__m512i) __builtin_ia32_blendmd_512_mask ((__v16si) __A,
          (__v16si) __W,
          (__mmask16) __U);
}
# 10747 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sqrt_pd (__m512d __A)
{
  return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
        (__v8df)
        _mm512_undefined_pd (),
        (__mmask8) -1,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sqrt_pd (__m512d __W, __mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
        (__v8df) __W,
        (__mmask8) __U,
        0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sqrt_pd (__mmask8 __U, __m512d __A)
{
  return (__m512d) __builtin_ia32_sqrtpd512_mask ((__v8df) __A,
        (__v8df)
        _mm512_setzero_pd (),
        (__mmask8) __U,
        0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sqrt_ps (__m512 __A)
{
  return (__m512) __builtin_ia32_sqrtps512_mask ((__v16sf) __A,
       (__v16sf)
       _mm512_undefined_ps (),
       (__mmask16) -1,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sqrt_ps (__m512 __W, __mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_sqrtps512_mask ((__v16sf) __A,
       (__v16sf) __W,
       (__mmask16) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sqrt_ps (__mmask16 __U, __m512 __A)
{
  return (__m512) __builtin_ia32_sqrtps512_mask ((__v16sf) __A,
       (__v16sf)
       _mm512_setzero_ps (),
       (__mmask16) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_pd (__m512d __A, __m512d __B)
{
  return (__m512d) ((__v8df)__A + (__v8df)__B);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_addpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_addpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_ps (__m512 __A, __m512 __B)
{
  return (__m512) ((__v16sf)__A + (__v16sf)__B);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_addps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_addps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_pd (__m512d __A, __m512d __B)
{
  return (__m512d) ((__v8df)__A - (__v8df)__B);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_subpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_subpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_ps (__m512 __A, __m512 __B)
{
  return (__m512) ((__v16sf)__A - (__v16sf)__B);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_subps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_subps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mul_pd (__m512d __A, __m512d __B)
{
  return (__m512d) ((__v8df)__A * (__v8df)__B);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mul_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_mulpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mul_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_mulpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mul_ps (__m512 __A, __m512 __B)
{
  return (__m512) ((__v16sf)__A * (__v16sf)__B);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mul_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_mulps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mul_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_mulps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_div_pd (__m512d __M, __m512d __V)
{
  return (__m512d) ((__v8df)__M / (__v8df)__V);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_div_pd (__m512d __W, __mmask8 __U, __m512d __M, __m512d __V)
{
  return (__m512d) __builtin_ia32_divpd512_mask ((__v8df) __M,
       (__v8df) __V,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_div_pd (__mmask8 __U, __m512d __M, __m512d __V)
{
  return (__m512d) __builtin_ia32_divpd512_mask ((__v8df) __M,
       (__v8df) __V,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_div_ps (__m512 __A, __m512 __B)
{
  return (__m512) ((__v16sf)__A / (__v16sf)__B);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_div_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_divps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_div_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_divps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_pd (__m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_undefined_pd (),
       (__mmask8) -1,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_maxpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_ps (__m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_maxps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_undefined_ps (),
      (__mmask16) -1,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_maxps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_maxps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_pd (__m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_undefined_pd (),
       (__mmask8) -1,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __W,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_minpd512_mask ((__v8df) __A,
       (__v8df) __B,
       (__v8df)
       _mm512_setzero_pd (),
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_ps (__m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_minps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_undefined_ps (),
      (__mmask16) -1,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_minps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf) __W,
      (__mmask16) __U,
      0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_minps512_mask ((__v16sf) __A,
      (__v16sf) __B,
      (__v16sf)
      _mm512_setzero_ps (),
      (__mmask16) __U,
      0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_scalef_pd (__m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_scalefpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_undefined_pd (),
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_scalef_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_scalefpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df) __W,
          (__mmask8) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_scalef_pd (__mmask8 __U, __m512d __A, __m512d __B)
{
  return (__m512d) __builtin_ia32_scalefpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df)
          _mm512_setzero_pd (),
          (__mmask8) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_scalef_ps (__m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_scalefps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_scalef_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_scalefps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __W,
         (__mmask16) __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_scalef_ps (__mmask16 __U, __m512 __A, __m512 __B)
{
  return (__m512) __builtin_ia32_scalefps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U,
         0x04);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_scalef_sd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_scalefsd_round ((__v2df) __A,
        (__v2df) __B,
        0x04);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_scalef_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_scalefss_round ((__v4sf) __A,
       (__v4sf) __B,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmadd_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df) __C,
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmadd_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask ((__v8df) __A,
          (__v8df) __B,
          (__v8df) __C,
          (__mmask8) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmadd_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask3 ((__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmadd_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_maskz ((__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmadd_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __C,
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmadd_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __C,
         (__mmask16) __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmadd_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask3 ((__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmadd_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_maskz ((__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmsub_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask ((__v8df) __A,
          (__v8df) __B,
          -(__v8df) __C,
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmsub_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask ((__v8df) __A,
          (__v8df) __B,
          -(__v8df) __C,
          (__mmask8) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmsub_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfmsubpd512_mask3 ((__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmsub_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_maskz ((__v8df) __A,
           (__v8df) __B,
           -(__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmsub_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         -(__v16sf) __C,
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmsub_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask ((__v16sf) __A,
         (__v16sf) __B,
         -(__v16sf) __C,
         (__mmask16) __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmsub_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfmsubps512_mask3 ((__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmsub_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_maskz ((__v16sf) __A,
          (__v16sf) __B,
          -(__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmaddsub_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_mask ((__v8df) __A,
             (__v8df) __B,
             (__v8df) __C,
             (__mmask8) -1,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmaddsub_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_mask ((__v8df) __A,
             (__v8df) __B,
             (__v8df) __C,
             (__mmask8) __U,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmaddsub_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_mask3 ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __C,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmaddsub_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_maskz ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __C,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmaddsub_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_mask ((__v16sf) __A,
            (__v16sf) __B,
            (__v16sf) __C,
            (__mmask16) -1,
            0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmaddsub_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_mask ((__v16sf) __A,
            (__v16sf) __B,
            (__v16sf) __C,
            (__mmask16) __U,
            0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmaddsub_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_mask3 ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf) __C,
             (__mmask16) __U,
             0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmaddsub_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_maskz ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf) __C,
             (__mmask16) __U,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmsubadd_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_mask ((__v8df) __A,
             (__v8df) __B,
             -(__v8df) __C,
             (__mmask8) -1,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmsubadd_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_mask ((__v8df) __A,
             (__v8df) __B,
             -(__v8df) __C,
             (__mmask8) __U,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmsubadd_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfmsubaddpd512_mask3 ((__v8df) __A,
       (__v8df) __B,
       (__v8df) __C,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmsubadd_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddsubpd512_maskz ((__v8df) __A,
       (__v8df) __B,
       -(__v8df) __C,
       (__mmask8) __U,
       0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fmsubadd_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_mask ((__v16sf) __A,
            (__v16sf) __B,
            -(__v16sf) __C,
            (__mmask16) -1,
            0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fmsubadd_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_mask ((__v16sf) __A,
            (__v16sf) __B,
            -(__v16sf) __C,
            (__mmask16) __U,
            0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fmsubadd_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfmsubaddps512_mask3 ((__v16sf) __A,
             (__v16sf) __B,
             (__v16sf) __C,
             (__mmask16) __U,
             0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fmsubadd_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddsubps512_maskz ((__v16sf) __A,
             (__v16sf) __B,
             -(__v16sf) __C,
             (__mmask16) __U,
             0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fnmadd_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask (-(__v8df) __A,
          (__v8df) __B,
          (__v8df) __C,
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fnmadd_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfnmaddpd512_mask ((__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fnmadd_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask3 (-(__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fnmadd_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_maskz (-(__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fnmadd_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask (-(__v16sf) __A,
         (__v16sf) __B,
         (__v16sf) __C,
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fnmadd_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfnmaddps512_mask ((__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fnmadd_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask3 (-(__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fnmadd_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_maskz (-(__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fnmsub_pd (__m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_mask (-(__v8df) __A,
          (__v8df) __B,
          -(__v8df) __C,
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fnmsub_pd (__m512d __A, __mmask8 __U, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfnmsubpd512_mask ((__v8df) __A,
           (__v8df) __B,
           (__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fnmsub_pd (__m512d __A, __m512d __B, __m512d __C, __mmask8 __U)
{
  return (__m512d) __builtin_ia32_vfnmsubpd512_mask3 ((__v8df) __A,
            (__v8df) __B,
            (__v8df) __C,
            (__mmask8) __U,
            0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fnmsub_pd (__mmask8 __U, __m512d __A, __m512d __B, __m512d __C)
{
  return (__m512d) __builtin_ia32_vfmaddpd512_maskz (-(__v8df) __A,
           (__v8df) __B,
           -(__v8df) __C,
           (__mmask8) __U,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_fnmsub_ps (__m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_mask (-(__v16sf) __A,
         (__v16sf) __B,
         -(__v16sf) __C,
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_fnmsub_ps (__m512 __A, __mmask16 __U, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfnmsubps512_mask ((__v16sf) __A,
          (__v16sf) __B,
          (__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask3_fnmsub_ps (__m512 __A, __m512 __B, __m512 __C, __mmask16 __U)
{
  return (__m512) __builtin_ia32_vfnmsubps512_mask3 ((__v16sf) __A,
           (__v16sf) __B,
           (__v16sf) __C,
           (__mmask16) __U,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_fnmsub_ps (__mmask16 __U, __m512 __A, __m512 __B, __m512 __C)
{
  return (__m512) __builtin_ia32_vfmaddps512_maskz (-(__v16sf) __A,
          (__v16sf) __B,
          -(__v16sf) __C,
          (__mmask16) __U,
          0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvttpd_epi32 (__m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2dq512_mask ((__v8df) __A,
           (__v8si)
           _mm256_undefined_si256 (),
           (__mmask8) -1,
           0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvttpd_epi32 (__m256i __W, __mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2dq512_mask ((__v8df) __A,
           (__v8si) __W,
           (__mmask8) __U,
           0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvttpd_epi32 (__mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2dq512_mask ((__v8df) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U,
           0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvttpd_epu32 (__m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2udq512_mask ((__v8df) __A,
            (__v8si)
            _mm256_undefined_si256 (),
            (__mmask8) -1,
            0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvttpd_epu32 (__m256i __W, __mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2udq512_mask ((__v8df) __A,
            (__v8si) __W,
            (__mmask8) __U,
            0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvttpd_epu32 (__mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvttpd2udq512_mask ((__v8df) __A,
            (__v8si)
            _mm256_setzero_si256 (),
            (__mmask8) __U,
            0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtpd_epi32 (__m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2dq512_mask ((__v8df) __A,
          (__v8si)
          _mm256_undefined_si256 (),
          (__mmask8) -1,
          0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtpd_epi32 (__m256i __W, __mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2dq512_mask ((__v8df) __A,
          (__v8si) __W,
          (__mmask8) __U,
          0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtpd_epi32 (__mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2dq512_mask ((__v8df) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U,
          0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtpd_epu32 (__m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2udq512_mask ((__v8df) __A,
           (__v8si)
           _mm256_undefined_si256 (),
           (__mmask8) -1,
           0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtpd_epu32 (__m256i __W, __mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2udq512_mask ((__v8df) __A,
           (__v8si) __W,
           (__mmask8) __U,
           0x04);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtpd_epu32 (__mmask8 __U, __m512d __A)
{
  return (__m256i) __builtin_ia32_cvtpd2udq512_mask ((__v8df) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvttps_epi32 (__m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2dq512_mask ((__v16sf) __A,
           (__v16si)
           _mm512_undefined_epi32 (),
           (__mmask16) -1,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvttps_epi32 (__m512i __W, __mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2dq512_mask ((__v16sf) __A,
           (__v16si) __W,
           (__mmask16) __U,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvttps_epi32 (__mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2dq512_mask ((__v16sf) __A,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvttps_epu32 (__m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2udq512_mask ((__v16sf) __A,
            (__v16si)
            _mm512_undefined_epi32 (),
            (__mmask16) -1,
            0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvttps_epu32 (__m512i __W, __mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2udq512_mask ((__v16sf) __A,
            (__v16si) __W,
            (__mmask16) __U,
            0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvttps_epu32 (__mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvttps2udq512_mask ((__v16sf) __A,
            (__v16si)
            _mm512_setzero_si512 (),
            (__mmask16) __U,
            0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtps_epi32 (__m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2dq512_mask ((__v16sf) __A,
          (__v16si)
          _mm512_undefined_epi32 (),
          (__mmask16) -1,
          0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtps_epi32 (__m512i __W, __mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2dq512_mask ((__v16sf) __A,
          (__v16si) __W,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtps_epi32 (__mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2dq512_mask ((__v16sf) __A,
          (__v16si)
          _mm512_setzero_si512 (),
          (__mmask16) __U,
          0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtps_epu32 (__m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2udq512_mask ((__v16sf) __A,
           (__v16si)
           _mm512_undefined_epi32 (),
           (__mmask16) -1,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtps_epu32 (__m512i __W, __mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2udq512_mask ((__v16sf) __A,
           (__v16si) __W,
           (__mmask16) __U,
           0x04);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtps_epu32 (__mmask16 __U, __m512 __A)
{
  return (__m512i) __builtin_ia32_cvtps2udq512_mask ((__v16sf) __A,
           (__v16si)
           _mm512_setzero_si512 (),
           (__mmask16) __U,
           0x04);
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsd_f64 (__m512d __A)
{
  return __A[0];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtss_f32 (__m512 __A)
{
  return __A[0];
}


extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtu64_ss (__m128 __A, unsigned long long __B)
{
  return (__m128) __builtin_ia32_cvtusi2ss64 ((__v4sf) __A, __B,
           0x04);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtu64_sd (__m128d __A, unsigned long long __B)
{
  return (__m128d) __builtin_ia32_cvtusi2sd64 ((__v2df) __A, __B,
            0x04);
}


extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtu32_ss (__m128 __A, unsigned __B)
{
  return (__m128) __builtin_ia32_cvtusi2ss32 ((__v4sf) __A, __B,
           0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi32_ps (__m512i __A)
{
  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
         (__v16sf)
         _mm512_undefined_ps (),
         (__mmask16) -1,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi32_ps (__m512 __W, __mmask16 __U, __m512i __A)
{
  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
         (__v16sf) __W,
         (__mmask16) __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi32_ps (__mmask16 __U, __m512i __A)
{
  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
         (__v16sf)
         _mm512_setzero_ps (),
         (__mmask16) __U,
         0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu32_ps (__m512i __A)
{
  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
          (__v16sf)
          _mm512_undefined_ps (),
          (__mmask16) -1,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu32_ps (__m512 __W, __mmask16 __U, __m512i __A)
{
  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
          (__v16sf) __W,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu32_ps (__mmask16 __U, __m512i __A)
{
  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
          (__v16sf)
          _mm512_setzero_ps (),
          (__mmask16) __U,
          0x04);
}
# 12377 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_u64 (__m128 __A)
{
  return (unsigned long long) __builtin_ia32_vcvtss2usi64 ((__v4sf)
          __A,
          0x04);
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_u64 (__m128 __A)
{
  return (unsigned long long) __builtin_ia32_vcvttss2usi64 ((__v4sf)
           __A,
           0x04);
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_i64 (__m128 __A)
{
  return (long long) __builtin_ia32_vcvttss2si64 ((__v4sf) __A,
        0x04);
}


extern __inline unsigned
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_u32 (__m128 __A)
{
  return (unsigned) __builtin_ia32_vcvtss2usi32 ((__v4sf) __A,
       0x04);
}

extern __inline unsigned
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_u32 (__m128 __A)
{
  return (unsigned) __builtin_ia32_vcvttss2usi32 ((__v4sf) __A,
        0x04);
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_i32 (__m128 __A)
{
  return (int) __builtin_ia32_vcvttss2si32 ((__v4sf) __A,
         0x04);
}


extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_u64 (__m128d __A)
{
  return (unsigned long long) __builtin_ia32_vcvtsd2usi64 ((__v2df)
          __A,
          0x04);
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_u64 (__m128d __A)
{
  return (unsigned long long) __builtin_ia32_vcvttsd2usi64 ((__v2df)
           __A,
           0x04);
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_i64 (__m128d __A)
{
  return (long long) __builtin_ia32_vcvttsd2si64 ((__v2df) __A,
        0x04);
}


extern __inline unsigned
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsd_u32 (__m128d __A)
{
  return (unsigned) __builtin_ia32_vcvtsd2usi32 ((__v2df) __A,
       0x04);
}

extern __inline unsigned
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_u32 (__m128d __A)
{
  return (unsigned) __builtin_ia32_vcvttsd2usi32 ((__v2df) __A,
        0x04);
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttsd_i32 (__m128d __A)
{
  return (int) __builtin_ia32_vcvttsd2si32 ((__v2df) __A,
         0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtps_pd (__m256 __A)
{
  return (__m512d) __builtin_ia32_cvtps2pd512_mask ((__v8sf) __A,
          (__v8df)
          _mm512_undefined_pd (),
          (__mmask8) -1,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtps_pd (__m512d __W, __mmask8 __U, __m256 __A)
{
  return (__m512d) __builtin_ia32_cvtps2pd512_mask ((__v8sf) __A,
          (__v8df) __W,
          (__mmask8) __U,
          0x04);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtps_pd (__mmask8 __U, __m256 __A)
{
  return (__m512d) __builtin_ia32_cvtps2pd512_mask ((__v8sf) __A,
          (__v8df)
          _mm512_setzero_pd (),
          (__mmask8) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtph_ps (__m256i __A)
{
  return (__m512) __builtin_ia32_vcvtph2ps512_mask ((__v16hi) __A,
          (__v16sf)
          _mm512_undefined_ps (),
          (__mmask16) -1,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtph_ps (__m512 __W, __mmask16 __U, __m256i __A)
{
  return (__m512) __builtin_ia32_vcvtph2ps512_mask ((__v16hi) __A,
          (__v16sf) __W,
          (__mmask16) __U,
          0x04);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtph_ps (__mmask16 __U, __m256i __A)
{
  return (__m512) __builtin_ia32_vcvtph2ps512_mask ((__v16hi) __A,
          (__v16sf)
          _mm512_setzero_ps (),
          (__mmask16) __U,
          0x04);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtpd_ps (__m512d __A)
{
  return (__m256) __builtin_ia32_cvtpd2ps512_mask ((__v8df) __A,
         (__v8sf)
         _mm256_undefined_ps (),
         (__mmask8) -1,
         0x04);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtpd_ps (__m256 __W, __mmask8 __U, __m512d __A)
{
  return (__m256) __builtin_ia32_cvtpd2ps512_mask ((__v8df) __A,
         (__v8sf) __W,
         (__mmask8) __U,
         0x04);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtpd_ps (__mmask8 __U, __m512d __A)
{
  return (__m256) __builtin_ia32_cvtpd2ps512_mask ((__v8df) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U,
         0x04);
}
# 13084 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kmov (__mmask16 __A)
{
  return __builtin_ia32_kmovw (__A);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd_ps (__m512d __A)
{
  return (__m512) (__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd_si512 (__m512d __A)
{
  return (__m512i) (__A);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps_pd (__m512 __A)
{
  return (__m512d) (__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps_si512 (__m512 __A)
{
  return (__m512i) (__A);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi512_ps (__m512i __A)
{
  return (__m512) (__A);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi512_pd (__m512i __A)
{
  return (__m512d) (__A);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd512_pd128 (__m512d __A)
{
  return (__m128d)((__m128) __builtin_ia32_extractf32x4_mask ((__v16sf)(__m512) ((__m512)__A), (int) (0), (__v4sf)(__m128)_mm_undefined_ps(), (__mmask8)-1));
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps512_ps128 (__m512 __A)
{
  return ((__m128) __builtin_ia32_extractf32x4_mask ((__v16sf)(__m512) (__A), (int) (0), (__v4sf)(__m128)_mm_undefined_ps(), (__mmask8)-1));
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi512_si128 (__m512i __A)
{
  return (__m128i)((__m128i) __builtin_ia32_extracti32x4_mask ((__v16si)(__m512i) ((__m512i)__A), (int) (0), (__v4si)(__m128i)_mm_undefined_si128 (), (__mmask8)-1));
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd512_pd256 (__m512d __A)
{
  return ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1));
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps512_ps256 (__m512 __A)
{
  return (__m256)((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d)__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1));
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi512_si256 (__m512i __A)
{
  return (__m256i)((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d)__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1));
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd128_pd512 (__m128d __A)
{
  return (__m512d) __builtin_ia32_pd512_pd((__m128d)__A);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps128_ps512 (__m128 __A)
{
  return (__m512) __builtin_ia32_ps512_ps((__m128)__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi128_si512 (__m128i __A)
{
  return (__m512i) __builtin_ia32_si512_si((__v4si)__A);
}

extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castpd256_pd512 (__m256d __A)
{
  return __builtin_ia32_pd512_256pd (__A);
}

extern __inline __m512
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castps256_ps512 (__m256 __A)
{
  return __builtin_ia32_ps512_256ps (__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_castsi256_si512 (__m256i __A)
{
  return (__m512i)__builtin_ia32_si512_256si ((__v8si)__A);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epu32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
           (__v16si) __B, 0,
           (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epu32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
           (__v16si) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epu64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
          (__v8di) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epu64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
          (__v8di) __B, 0,
          (__mmask8) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epu32_mask (__m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
           (__v16si) __B, 6,
           (__mmask16) -1);
}

extern __inline __mmask16
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epu32_mask (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__mmask16) __builtin_ia32_ucmpd512_mask ((__v16si) __A,
           (__v16si) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epu64_mask (__mmask8 __U, __m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
          (__v8di) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epu64_mask (__m512i __A, __m512i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq512_mask ((__v8di) __A,
          (__v8di) __B, 6,
          (__mmask8) -1);
}
# 13297 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_add_epi32 (__m512i __A)
{
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 + __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 + __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 + __T7; return __T8[0] + __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_mul_epi32 (__m512i __A)
{
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 * __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 * __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 * __T7; return __T8[0] * __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_and_epi32 (__m512i __A)
{
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 & __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 & __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 & __T7; return __T8[0] & __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_or_epi32 (__m512i __A)
{
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 | __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 | __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 | __T7; return __T8[0] | __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_add_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi32 (__U, __A);
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 + __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 + __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 + __T7; return __T8[0] + __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_mul_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi32 (_mm512_set1_epi32 (1), __U, __A);
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 * __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 * __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 * __T7; return __T8[0] * __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_and_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi32 (_mm512_set1_epi32 (~0), __U, __A);
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 & __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 & __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 & __T7; return __T8[0] & __T8[1];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_or_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi32 (__U, __A);
  __v8si __T1 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v8si __T2 = (__v8si) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 | __T2); __v4si __T4 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v4si __T5 = (__v4si) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v4si __T6 = __T4 | __T5; __v4si __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __v4si __T8 = __T6 | __T7; return __T8[0] | __T8[1];
}
# 13373 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_epi32 (__m512i __A)
{
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_min_epi32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_min_epi32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_min_epi32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_min_epi32 (__T8, __T9); return __T10[0];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_epi32 (__m512i __A)
{
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_max_epi32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_max_epi32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_max_epi32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_max_epi32 (__T8, __T9); return __T10[0];
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_epu32 (__m512i __A)
{
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_min_epu32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_min_epu32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_min_epu32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_min_epu32 (__T8, __T9); return __T10[0];
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_epu32 (__m512i __A)
{
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_max_epu32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_max_epu32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_max_epu32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_max_epu32 (__T8, __T9); return __T10[0];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi32 (_mm512_set1_epi32 (0x7fffffff), __U, __A);
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_min_epi32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_min_epi32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_min_epi32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_min_epi32 (__T8, __T9); return __T10[0];
}

extern __inline int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_epi32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi32 (_mm512_set1_epi32 (-0x7fffffff - 1), __U, __A);
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_max_epi32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_max_epi32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_max_epi32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_max_epi32 (__T8, __T9); return __T10[0];
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_epu32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi32 (_mm512_set1_epi32 (~0), __U, __A);
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_min_epu32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_min_epu32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_min_epu32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_min_epu32 (__T8, __T9); return __T10[0];
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_epu32 (__mmask16 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi32 (__U, __A);
  __m256i __T1 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T2 = (__m256i) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = _mm256_max_epu32 (__T1, __T2); __m128i __T4 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __m128i __T5 = (__m128i) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __m128i __T6 = _mm_max_epu32 (__T4, __T5); __m128i __T7 = (__m128i) __builtin_shuffle ((__v4si) __T6, (__v4si) { 2, 3, 0, 1 }); __m128i __T8 = _mm_max_epu32 (__T6, __T7); __m128i __T9 = (__m128i) __builtin_shuffle ((__v4si) __T8, (__v4si) { 1, 0, 1, 0 }); __v4si __T10 = (__v4si) _mm_max_epu32 (__T8, __T9); return __T10[0];
}
# 13445 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_add_ps (__m512 __A)
{
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = __T1 + __T2; __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = __T4 + __T5; __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = __T6 + __T7; return __T8[0] + __T8[1];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_mul_ps (__m512 __A)
{
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = __T1 * __T2; __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = __T4 * __T5; __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = __T6 * __T7; return __T8[0] * __T8[1];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_add_ps (__mmask16 __U, __m512 __A)
{
  __A = _mm512_maskz_mov_ps (__U, __A);
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = __T1 + __T2; __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = __T4 + __T5; __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = __T6 + __T7; return __T8[0] + __T8[1];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_mul_ps (__mmask16 __U, __m512 __A)
{
  __A = _mm512_mask_mov_ps (_mm512_set1_ps (1.0f), __U, __A);
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = __T1 * __T2; __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = __T4 * __T5; __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = __T6 * __T7; return __T8[0] * __T8[1];
}
# 13489 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_ps (__m512 __A)
{
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = _mm256_min_ps (__T1, __T2); __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = _mm_min_ps (__T4, __T5); __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = _mm_min_ps (__T6, __T7); __m128 __T9 = __builtin_shuffle (__T8, (__v4si) { 1, 0, 1, 0 }); __m128 __T10 = _mm_min_ps (__T8, __T9); return __T10[0];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_ps (__m512 __A)
{
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = _mm256_max_ps (__T1, __T2); __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = _mm_max_ps (__T4, __T5); __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = _mm_max_ps (__T6, __T7); __m128 __T9 = __builtin_shuffle (__T8, (__v4si) { 1, 0, 1, 0 }); __m128 __T10 = _mm_max_ps (__T8, __T9); return __T10[0];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_ps (__mmask16 __U, __m512 __A)
{
  __A = _mm512_mask_mov_ps (_mm512_set1_ps (__builtin_inff ()), __U, __A);
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = _mm256_min_ps (__T1, __T2); __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = _mm_min_ps (__T4, __T5); __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = _mm_min_ps (__T6, __T7); __m128 __T9 = __builtin_shuffle (__T8, (__v4si) { 1, 0, 1, 0 }); __m128 __T10 = _mm_min_ps (__T8, __T9); return __T10[0];
}

extern __inline float
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_ps (__mmask16 __U, __m512 __A)
{
  __A = _mm512_mask_mov_ps (_mm512_set1_ps (-__builtin_inff ()), __U, __A);
  __m256 __T1 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T2 = (__m256) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) ((__m512d) __A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256 __T3 = _mm256_max_ps (__T1, __T2); __m128 __T4 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(1))); __m128 __T5 = ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(__T3), (int)(0))); __m128 __T6 = _mm_max_ps (__T4, __T5); __m128 __T7 = __builtin_shuffle (__T6, (__v4si) { 2, 3, 0, 1 }); __m128 __T8 = _mm_max_ps (__T6, __T7); __m128 __T9 = __builtin_shuffle (__T8, (__v4si) { 1, 0, 1, 0 }); __m128 __T10 = _mm_max_ps (__T8, __T9); return __T10[0];
}
# 13529 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_add_epi64 (__m512i __A)
{
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 + __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 + __T5; return __T6[0] + __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_mul_epi64 (__m512i __A)
{
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 * __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 * __T5; return __T6[0] * __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_and_epi64 (__m512i __A)
{
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 & __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 & __T5; return __T6[0] & __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_or_epi64 (__m512i __A)
{
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 | __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 | __T5; return __T6[0] | __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_add_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi64 (__U, __A);
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 + __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 + __T5; return __T6[0] + __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_mul_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi64 (_mm512_set1_epi64 (1LL), __U, __A);
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 * __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 * __T5; return __T6[0] * __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_and_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi64 (_mm512_set1_epi64 (~0LL), __U, __A);
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 & __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 & __T5; return __T6[0] & __T6[1];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_or_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi64 (__U, __A);
  __v4di __T1 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (1), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __v4di __T2 = (__v4di) ((__m256i) __builtin_ia32_extracti64x4_mask ((__v8di)(__m512i) (__A), (int) (0), (__v4di)(__m256i)_mm256_undefined_si256 (), (__mmask8)-1)); __m256i __T3 = (__m256i) (__T1 | __T2); __v2di __T4 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(1))); __v2di __T5 = (__v2di) ((__m128i) __builtin_ia32_extract128i256 ((__v4di)(__m256i)(__T3), (int)(0))); __v2di __T6 = __T4 | __T5; return __T6[0] | __T6[1];
}
# 13603 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_epi64 (__m512i __A)
{
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_min_epi64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_min_epi64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_min_epi64 (__T4, __T5); return __T6[0];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_epi64 (__m512i __A)
{
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_max_epi64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_max_epi64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_max_epi64 (__T4, __T5); return __T6[0];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi64 (_mm512_set1_epi64 (0x7fffffffffffffffLL),
          __U, __A);
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_min_epi64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_min_epi64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_min_epi64 (__T4, __T5); return __T6[0];
}

extern __inline long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_epi64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi64 (_mm512_set1_epi64 (-0x7fffffffffffffffLL - 1),
          __U, __A);
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_max_epi64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_max_epi64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_max_epi64 (__T4, __T5); return __T6[0];
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_epu64 (__m512i __A)
{
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_min_epu64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_min_epu64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_min_epu64 (__T4, __T5); return __T6[0];
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_epu64 (__m512i __A)
{
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_max_epu64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_max_epu64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_max_epu64 (__T4, __T5); return __T6[0];
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_epu64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_mask_mov_epi64 (_mm512_set1_epi64 (~0LL), __U, __A);
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_min_epu64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_min_epu64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_min_epu64 (__T4, __T5); return __T6[0];
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_epu64 (__mmask8 __U, __m512i __A)
{
  __A = _mm512_maskz_mov_epi64 (__U, __A);
  __m512i __T1 = ((__m512i) __builtin_ia32_shuf_i64x2_mask ((__v8di)(__m512i)(__A), (__v8di)(__m512i)(__A), (int)(0x4e), (__v8di)(__m512i)_mm512_undefined_epi32 (), (__mmask8)-1)); __m512i __T2 = _mm512_max_epu64 (__A, __T1); __m512i __T3 = (__m512i) __builtin_shuffle ((__v8di) __T2, (__v8di) { 2, 3, 0, 1, 6, 7, 4, 5 }); __m512i __T4 = _mm512_max_epu64 (__T2, __T3); __m512i __T5 = (__m512i) __builtin_shuffle ((__v8di) __T4, (__v8di) { 1, 0, 3, 2, 5, 4, 7, 6 }); __v8di __T6 = (__v8di) _mm512_max_epu64 (__T4, __T5); return __T6[0];
}
# 13675 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_add_pd (__m512d __A)
{
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = __T1 + __T2; __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = __T4 + __T5; return __T6[0] + __T6[1];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_mul_pd (__m512d __A)
{
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = __T1 * __T2; __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = __T4 * __T5; return __T6[0] * __T6[1];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_add_pd (__mmask8 __U, __m512d __A)
{
  __A = _mm512_maskz_mov_pd (__U, __A);
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = __T1 + __T2; __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = __T4 + __T5; return __T6[0] + __T6[1];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_mul_pd (__mmask8 __U, __m512d __A)
{
  __A = _mm512_mask_mov_pd (_mm512_set1_pd (1.0), __U, __A);
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = __T1 * __T2; __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = __T4 * __T5; return __T6[0] * __T6[1];
}
# 13717 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512fintrin.h" 3 4
extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_min_pd (__m512d __A)
{
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = _mm256_min_pd (__T1, __T2); __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = _mm_min_pd (__T4, __T5); __m128d __T7 = (__m128d) __builtin_shuffle (__T6, (__v2di) { 1, 0 }); __m128d __T8 = _mm_min_pd (__T6, __T7); return __T8[0];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_reduce_max_pd (__m512d __A)
{
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = _mm256_max_pd (__T1, __T2); __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = _mm_max_pd (__T4, __T5); __m128d __T7 = (__m128d) __builtin_shuffle (__T6, (__v2di) { 1, 0 }); __m128d __T8 = _mm_max_pd (__T6, __T7); return __T8[0];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_min_pd (__mmask8 __U, __m512d __A)
{
  __A = _mm512_mask_mov_pd (_mm512_set1_pd (__builtin_inf ()), __U, __A);
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = _mm256_min_pd (__T1, __T2); __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = _mm_min_pd (__T4, __T5); __m128d __T7 = (__m128d) __builtin_shuffle (__T6, (__v2di) { 1, 0 }); __m128d __T8 = _mm_min_pd (__T6, __T7); return __T8[0];
}

extern __inline double
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_reduce_max_pd (__mmask8 __U, __m512d __A)
{
  __A = _mm512_mask_mov_pd (_mm512_set1_pd (-__builtin_inf ()), __U, __A);
  __m256d __T1 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (1), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T2 = (__m256d) ((__m256d) __builtin_ia32_extractf64x4_mask ((__v8df)(__m512d) (__A), (int) (0), (__v4df)(__m256d)_mm256_undefined_pd(), (__mmask8)-1)); __m256d __T3 = _mm256_max_pd (__T1, __T2); __m128d __T4 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(1))); __m128d __T5 = ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(__T3), (int)(0))); __m128d __T6 = _mm_max_pd (__T4, __T5); __m128d __T7 = (__m128d) __builtin_shuffle (__T6, (__v2di) { 1, 0 }); __m128d __T8 = _mm_max_pd (__T6, __T7); return __T8[0];
}





#pragma GCC pop_options
# 46 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512erintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512erintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512er")




typedef double __v8df __attribute__ ((__vector_size__ (64)));
typedef float __v16sf __attribute__ ((__vector_size__ (64)));



typedef float __m512 __attribute__ ((__vector_size__ (64), __may_alias__));
typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));

typedef unsigned char __mmask8;
typedef unsigned short __mmask16;
# 391 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512erintrin.h" 3 4
#pragma GCC pop_options
# 48 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512pfintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512pfintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512pf")




typedef long long __v8di __attribute__ ((__vector_size__ (64)));
typedef int __v16si __attribute__ ((__vector_size__ (64)));



typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));

typedef unsigned char __mmask8;
typedef unsigned short __mmask16;
# 261 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512pfintrin.h" 3 4
#pragma GCC pop_options
# 50 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512cdintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512cdintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512cd")




typedef long long __v8di __attribute__ ((__vector_size__ (64)));
typedef int __v16si __attribute__ ((__vector_size__ (64)));



typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
typedef double __m512d __attribute__ ((__vector_size__ (64), __may_alias__));

typedef unsigned char __mmask8;
typedef unsigned short __mmask16;

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_conflict_epi32 (__m512i __A)
{
  return (__m512i)
  __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
            (__v16si) _mm512_setzero_si512 (),
            (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_conflict_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
        (__v16si) __W,
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_conflict_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i)
  __builtin_ia32_vpconflictsi_512_mask ((__v16si) __A,
            (__v16si) _mm512_setzero_si512 (),
            (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_conflict_epi64 (__m512i __A)
{
  return (__m512i)
  __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
            (__v8di) _mm512_setzero_si512 (),
            (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_conflict_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
        (__v8di) __W,
        (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_conflict_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i)
  __builtin_ia32_vpconflictdi_512_mask ((__v8di) __A,
            (__v8di) _mm512_setzero_si512 (),
            (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_lzcnt_epi64 (__m512i __A)
{
  return (__m512i)
  __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
        (__v8di) _mm512_setzero_si512 (),
        (__mmask8) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_lzcnt_epi64 (__m512i __W, __mmask8 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
           (__v8di) __W,
           (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_lzcnt_epi64 (__mmask8 __U, __m512i __A)
{
  return (__m512i)
  __builtin_ia32_vplzcntq_512_mask ((__v8di) __A,
        (__v8di) _mm512_setzero_si512 (),
        (__mmask8) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_lzcnt_epi32 (__m512i __A)
{
  return (__m512i)
  __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
        (__v16si) _mm512_setzero_si512 (),
        (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_lzcnt_epi32 (__m512i __W, __mmask16 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
           (__v16si) __W,
           (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_lzcnt_epi32 (__mmask16 __U, __m512i __A)
{
  return (__m512i)
  __builtin_ia32_vplzcntd_512_mask ((__v16si) __A,
        (__v16si) _mm512_setzero_si512 (),
        (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastmb_epi64 (__mmask8 __A)
{
  return (__m512i) __builtin_ia32_broadcastmb512 (__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastmw_epi32 (__mmask16 __A)
{
  return (__m512i) __builtin_ia32_broadcastmw512 (__A);
}



#pragma GCC pop_options
# 52 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512vlintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512vlintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512vl")




typedef unsigned int __mmask32;

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mov_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_movapd256_mask ((__v4df) __A,
        (__v4df) __W,
        (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mov_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_movapd256_mask ((__v4df) __A,
        (__v4df)
        _mm256_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mov_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_movapd128_mask ((__v2df) __A,
        (__v2df) __W,
        (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mov_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_movapd128_mask ((__v2df) __A,
        (__v2df)
        _mm_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_load_pd (__m256d __W, __mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_loadapd256_mask ((__v4df *) __P,
         (__v4df) __W,
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_load_pd (__mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_loadapd256_mask ((__v4df *) __P,
         (__v4df)
         _mm256_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_load_pd (__m128d __W, __mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_loadapd128_mask ((__v2df *) __P,
         (__v2df) __W,
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_load_pd (__mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_loadapd128_mask ((__v2df *) __P,
         (__v2df)
         _mm_setzero_pd (),
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_store_pd (void *__P, __mmask8 __U, __m256d __A)
{
  __builtin_ia32_storeapd256_mask ((__v4df *) __P,
       (__v4df) __A,
       (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_store_pd (void *__P, __mmask8 __U, __m128d __A)
{
  __builtin_ia32_storeapd128_mask ((__v2df *) __P,
       (__v2df) __A,
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mov_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movaps256_mask ((__v8sf) __A,
       (__v8sf) __W,
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mov_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movaps256_mask ((__v8sf) __A,
       (__v8sf)
       _mm256_setzero_ps (),
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mov_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movaps128_mask ((__v4sf) __A,
       (__v4sf) __W,
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mov_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movaps128_mask ((__v4sf) __A,
       (__v4sf)
       _mm_setzero_ps (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_load_ps (__m256 __W, __mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_loadaps256_mask ((__v8sf *) __P,
        (__v8sf) __W,
        (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_load_ps (__mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_loadaps256_mask ((__v8sf *) __P,
        (__v8sf)
        _mm256_setzero_ps (),
        (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_load_ps (__m128 __W, __mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_loadaps128_mask ((__v4sf *) __P,
        (__v4sf) __W,
        (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_load_ps (__mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_loadaps128_mask ((__v4sf *) __P,
        (__v4sf)
        _mm_setzero_ps (),
        (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_store_ps (void *__P, __mmask8 __U, __m256 __A)
{
  __builtin_ia32_storeaps256_mask ((__v8sf *) __P,
       (__v8sf) __A,
       (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_store_ps (void *__P, __mmask8 __U, __m128 __A)
{
  __builtin_ia32_storeaps128_mask ((__v4sf *) __P,
       (__v4sf) __A,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mov_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_movdqa64_256_mask ((__v4di) __A,
           (__v4di) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mov_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_movdqa64_256_mask ((__v4di) __A,
           (__v4di)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mov_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_movdqa64_128_mask ((__v2di) __A,
           (__v2di) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mov_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_movdqa64_128_mask ((__v2di) __A,
           (__v2di)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_load_epi64 (__m256i __W, __mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_movdqa64load256_mask ((__v4di *) __P,
       (__v4di) __W,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_load_epi64 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_movdqa64load256_mask ((__v4di *) __P,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_load_epi64 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_movdqa64load128_mask ((__v2di *) __P,
       (__v2di) __W,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_load_epi64 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_movdqa64load128_mask ((__v2di *) __P,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8)
       __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_store_epi64 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_movdqa64store256_mask ((__v4di *) __P,
     (__v4di) __A,
     (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_store_epi64 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_movdqa64store128_mask ((__v2di *) __P,
     (__v2di) __A,
     (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mov_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_movdqa32_256_mask ((__v8si) __A,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mov_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_movdqa32_256_mask ((__v8si) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mov_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_movdqa32_128_mask ((__v4si) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mov_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_movdqa32_128_mask ((__v4si) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_load_epi32 (__m256i __W, __mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_movdqa32load256_mask ((__v8si *) __P,
       (__v8si) __W,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_load_epi32 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_movdqa32load256_mask ((__v8si *) __P,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_load_epi32 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_movdqa32load128_mask ((__v4si *) __P,
       (__v4si) __W,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_load_epi32 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_movdqa32load128_mask ((__v4si *) __P,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8)
       __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_store_epi32 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_movdqa32store256_mask ((__v8si *) __P,
     (__v8si) __A,
     (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_store_epi32 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_movdqa32store128_mask ((__v4si *) __P,
     (__v4si) __A,
     (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_add_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_addpd128_mask ((__v2df) __A,
       (__v2df) __B,
       (__v2df) __W,
       (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_add_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_addpd128_mask ((__v2df) __A,
       (__v2df) __B,
       (__v2df)
       _mm_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_add_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_addpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_add_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_addpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_add_ps (__m128 __W, __mmask16 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_addps128_mask ((__v4sf) __A,
      (__v4sf) __B,
      (__v4sf) __W,
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_add_ps (__mmask16 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_addps128_mask ((__v4sf) __A,
      (__v4sf) __B,
      (__v4sf)
      _mm_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_add_ps (__m256 __W, __mmask16 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_addps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_add_ps (__mmask16 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_addps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sub_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_subpd128_mask ((__v2df) __A,
       (__v2df) __B,
       (__v2df) __W,
       (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sub_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_subpd128_mask ((__v2df) __A,
       (__v2df) __B,
       (__v2df)
       _mm_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sub_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_subpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sub_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_subpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sub_ps (__m128 __W, __mmask16 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_subps128_mask ((__v4sf) __A,
      (__v4sf) __B,
      (__v4sf) __W,
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sub_ps (__mmask16 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_subps128_mask ((__v4sf) __A,
      (__v4sf) __B,
      (__v4sf)
      _mm_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sub_ps (__m256 __W, __mmask16 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_subps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sub_ps (__mmask16 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_subps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_store_epi64 (void *__P, __m256i __A)
{
  *(__m256i *) __P = __A;
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_epi64 (void *__P, __m128i __A)
{
  *(__m128i *) __P = __A;
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_loadu_pd (__m256d __W, __mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_loadupd256_mask ((const double *) __P,
         (__v4df) __W,
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_loadu_pd (__mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_loadupd256_mask ((const double *) __P,
         (__v4df)
         _mm256_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_loadu_pd (__m128d __W, __mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_loadupd128_mask ((const double *) __P,
         (__v2df) __W,
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_loadu_pd (__mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_loadupd128_mask ((const double *) __P,
         (__v2df)
         _mm_setzero_pd (),
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_storeu_pd (void *__P, __mmask8 __U, __m256d __A)
{
  __builtin_ia32_storeupd256_mask ((double *) __P,
       (__v4df) __A,
       (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_storeu_pd (void *__P, __mmask8 __U, __m128d __A)
{
  __builtin_ia32_storeupd128_mask ((double *) __P,
       (__v2df) __A,
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_loadu_ps (__m256 __W, __mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_loadups256_mask ((const float *) __P,
        (__v8sf) __W,
        (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_loadu_ps (__mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_loadups256_mask ((const float *) __P,
        (__v8sf)
        _mm256_setzero_ps (),
        (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_loadu_ps (__m128 __W, __mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_loadups128_mask ((const float *) __P,
        (__v4sf) __W,
        (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_loadu_ps (__mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_loadups128_mask ((const float *) __P,
        (__v4sf)
        _mm_setzero_ps (),
        (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_storeu_ps (void *__P, __mmask8 __U, __m256 __A)
{
  __builtin_ia32_storeups256_mask ((float *) __P,
       (__v8sf) __A,
       (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_storeu_ps (void *__P, __mmask8 __U, __m128 __A)
{
  __builtin_ia32_storeups128_mask ((float *) __P,
       (__v4sf) __A,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_loadu_epi64 (__m256i __W, __mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_loaddqudi256_mask ((const long long *) __P,
           (__v4di) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_loadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_loaddqudi256_mask ((const long long *) __P,
           (__v4di)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_loadu_epi64 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_loaddqudi128_mask ((const long long *) __P,
           (__v2di) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_loadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_loaddqudi128_mask ((const long long *) __P,
           (__v2di)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_storeu_epi64 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_storedqudi256_mask ((long long *) __P,
         (__v4di) __A,
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_storeu_epi64 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_storedqudi128_mask ((long long *) __P,
         (__v2di) __A,
         (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_loadu_epi32 (__m256i __W, __mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_loaddqusi256_mask ((const int *) __P,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_loadu_epi32 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_loaddqusi256_mask ((const int *) __P,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_loadu_epi32 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_loaddqusi128_mask ((const int *) __P,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_loadu_epi32 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_loaddqusi128_mask ((const int *) __P,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_storeu_epi32 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_storedqusi256_mask ((int *) __P,
         (__v8si) __A,
         (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_storeu_epi32 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_storedqusi128_mask ((int *) __P,
         (__v4si) __A,
         (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_abs_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_pabsd256_mask ((__v8si) __A,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_abs_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_pabsd256_mask ((__v8si) __A,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_abs_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pabsd128_mask ((__v4si) __A,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_abs_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pabsd128_mask ((__v4si) __A,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_abs_epi64 (__m256i __A)
{
  return (__m256i) __builtin_ia32_pabsq256_mask ((__v4di) __A,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_abs_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_pabsq256_mask ((__v4di) __A,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_abs_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_pabsq256_mask ((__v4di) __A,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_abs_epi64 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pabsq128_mask ((__v2di) __A,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_abs_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pabsq128_mask ((__v2di) __A,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_abs_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pabsq128_mask ((__v2di) __A,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtpd_epu32 (__m256d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq256_mask ((__v4df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtpd_epu32 (__m128i __W, __mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq256_mask ((__v4df) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtpd_epu32 (__mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq256_mask ((__v4df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpd_epu32 (__m128d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq128_mask ((__v2df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtpd_epu32 (__m128i __W, __mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq128_mask ((__v2df) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtpd_epu32 (__mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2udq128_mask ((__v2df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvttps_epi32 (__m256i __W, __mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvttps2dq256_mask ((__v8sf) __A,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvttps_epi32 (__mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvttps2dq256_mask ((__v8sf) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvttps_epi32 (__m128i __W, __mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvttps2dq128_mask ((__v4sf) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvttps_epi32 (__mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvttps2dq128_mask ((__v4sf) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvttps_epu32 (__m256 __A)
{
  return (__m256i) __builtin_ia32_cvttps2udq256_mask ((__v8sf) __A,
            (__v8si)
            _mm256_setzero_si256 (),
            (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvttps_epu32 (__m256i __W, __mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvttps2udq256_mask ((__v8sf) __A,
            (__v8si) __W,
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvttps_epu32 (__mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvttps2udq256_mask ((__v8sf) __A,
            (__v8si)
            _mm256_setzero_si256 (),
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttps_epu32 (__m128 __A)
{
  return (__m128i) __builtin_ia32_cvttps2udq128_mask ((__v4sf) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvttps_epu32 (__m128i __W, __mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvttps2udq128_mask ((__v4sf) __A,
            (__v4si) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvttps_epu32 (__mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvttps2udq128_mask ((__v4sf) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvttpd_epi32 (__m128i __W, __mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2dq256_mask ((__v4df) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvttpd_epi32 (__mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2dq256_mask ((__v4df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvttpd_epi32 (__m128i __W, __mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2dq128_mask ((__v2df) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvttpd_epi32 (__mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2dq128_mask ((__v2df) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvttpd_epu32 (__m256d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq256_mask ((__v4df) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvttpd_epu32 (__m128i __W, __mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq256_mask ((__v4df) __A,
            (__v4si) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvttpd_epu32 (__mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq256_mask ((__v4df) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttpd_epu32 (__m128d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq128_mask ((__v2df) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvttpd_epu32 (__m128i __W, __mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq128_mask ((__v2df) __A,
            (__v4si) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvttpd_epu32 (__mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvttpd2udq128_mask ((__v2df) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtpd_epi32 (__m128i __W, __mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2dq256_mask ((__v4df) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtpd_epi32 (__mmask8 __U, __m256d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2dq256_mask ((__v4df) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtpd_epi32 (__m128i __W, __mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2dq128_mask ((__v2df) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtpd_epi32 (__mmask8 __U, __m128d __A)
{
  return (__m128i) __builtin_ia32_cvtpd2dq128_mask ((__v2df) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_pd (__m256d __W, __mmask8 __U, __m128i __A)
{
  return (__m256d) __builtin_ia32_cvtdq2pd256_mask ((__v4si) __A,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi32_pd (__mmask8 __U, __m128i __A)
{
  return (__m256d) __builtin_ia32_cvtdq2pd256_mask ((__v4si) __A,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_pd (__m128d __W, __mmask8 __U, __m128i __A)
{
  return (__m128d) __builtin_ia32_cvtdq2pd128_mask ((__v4si) __A,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi32_pd (__mmask8 __U, __m128i __A)
{
  return (__m128d) __builtin_ia32_cvtdq2pd128_mask ((__v4si) __A,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu32_pd (__m128i __A)
{
  return (__m256d) __builtin_ia32_cvtudq2pd256_mask ((__v4si) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu32_pd (__m256d __W, __mmask8 __U, __m128i __A)
{
  return (__m256d) __builtin_ia32_cvtudq2pd256_mask ((__v4si) __A,
           (__v4df) __W,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu32_pd (__mmask8 __U, __m128i __A)
{
  return (__m256d) __builtin_ia32_cvtudq2pd256_mask ((__v4si) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu32_pd (__m128i __A)
{
  return (__m128d) __builtin_ia32_cvtudq2pd128_mask ((__v4si) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu32_pd (__m128d __W, __mmask8 __U, __m128i __A)
{
  return (__m128d) __builtin_ia32_cvtudq2pd128_mask ((__v4si) __A,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu32_pd (__mmask8 __U, __m128i __A)
{
  return (__m128d) __builtin_ia32_cvtudq2pd128_mask ((__v4si) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_ps (__m256 __W, __mmask8 __U, __m256i __A)
{
  return (__m256) __builtin_ia32_cvtdq2ps256_mask ((__v8si) __A,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi32_ps (__mmask16 __U, __m256i __A)
{
  return (__m256) __builtin_ia32_cvtdq2ps256_mask ((__v8si) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_ps (__m128 __W, __mmask8 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_cvtdq2ps128_mask ((__v4si) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi32_ps (__mmask16 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_cvtdq2ps128_mask ((__v4si) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepu32_ps (__m256i __A)
{
  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu32_ps (__m256 __W, __mmask8 __U, __m256i __A)
{
  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
          (__v8sf) __W,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu32_ps (__mmask8 __U, __m256i __A)
{
  return (__m256) __builtin_ia32_cvtudq2ps256_mask ((__v8si) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepu32_ps (__m128i __A)
{
  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu32_ps (__m128 __W, __mmask8 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu32_ps (__mmask8 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_cvtudq2ps128_mask ((__v4si) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtps_pd (__m256d __W, __mmask8 __U, __m128 __A)
{
  return (__m256d) __builtin_ia32_cvtps2pd256_mask ((__v4sf) __A,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtps_pd (__mmask8 __U, __m128 __A)
{
  return (__m256d) __builtin_ia32_cvtps2pd256_mask ((__v4sf) __A,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtps_pd (__m128d __W, __mmask8 __U, __m128 __A)
{
  return (__m128d) __builtin_ia32_cvtps2pd128_mask ((__v4sf) __A,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtps_pd (__mmask8 __U, __m128 __A)
{
  return (__m128d) __builtin_ia32_cvtps2pd128_mask ((__v4sf) __A,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi32_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovdb128mem_mask ((__v16qi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A,
        (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi32_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdb128_mask ((__v4si) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi32_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A,
        (__v16qi) __O, __M);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovdb256mem_mask ((__v16qi *) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi32_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsepi32_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb128_mask ((__v4si) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi32_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovsdb128mem_mask ((__v16qi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi32_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb128_mask ((__v4si) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtsepi32_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb128_mask ((__v4si) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsepi32_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb256_mask ((__v8si) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi32_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovsdb256mem_mask ((__v16qi *) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi32_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb256_mask ((__v8si) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtsepi32_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdb256_mask ((__v8si) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtusepi32_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb128_mask ((__v4si) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi32_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovusdb128mem_mask ((__v16qi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi32_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb128_mask ((__v4si) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtusepi32_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb128_mask ((__v4si) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtusepi32_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb256_mask ((__v8si) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi32_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovusdb256mem_mask ((__v16qi*) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi32_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb256_mask ((__v8si) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtusepi32_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdb256_mask ((__v8si) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi32_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovdw128mem_mask ((__v8hi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A,
        (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi32_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovdw128_mask ((__v4si) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi32_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovdw256mem_mask ((__v8hi *) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
        (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi32_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsepi32_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw128_mask ((__v4si) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovsdw128mem_mask ((__v8hi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi32_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw128_mask ((__v4si) __A,
         (__v8hi)__O,
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtsepi32_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw128_mask ((__v4si) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsepi32_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw256_mask ((__v8si) __A,
         (__v8hi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi32_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovsdw256mem_mask ((__v8hi *) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw256_mask ((__v8si) __A,
         (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtsepi32_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsdw256_mask ((__v8si) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtusepi32_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw128_mask ((__v4si) __A,
          (__v8hi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovusdw128mem_mask ((__v8hi *) __P, (__v4si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi32_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw128_mask ((__v4si) __A,
          (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtusepi32_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw128_mask ((__v4si) __A,
          (__v8hi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtusepi32_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw256_mask ((__v8si) __A,
          (__v8hi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi32_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovusdw256mem_mask ((__v8hi *) __P, (__v8si) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw256_mask ((__v8si) __A,
          (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtusepi32_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusdw256_mask ((__v8si) __A,
          (__v8hi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi64_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovqb128mem_mask ((__v16qi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A,
        (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi64_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqb128_mask ((__v2di) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi64_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A,
        (__v16qi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovqb256mem_mask ((__v16qi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A,
        (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi64_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqb256_mask ((__v4di) __A,
        (__v16qi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsepi64_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb128_mask ((__v2di) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovsqb128mem_mask ((__v16qi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb128_mask ((__v2di) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtsepi64_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb128_mask ((__v2di) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsepi64_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb256_mask ((__v4di) __A,
         (__v16qi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovsqb256mem_mask ((__v16qi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb256_mask ((__v4di) __A,
         (__v16qi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtsepi64_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqb256_mask ((__v4di) __A,
         (__v16qi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtusepi64_epi8 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb128_mask ((__v2di) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovusqb128mem_mask ((__v16qi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_epi8 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb128_mask ((__v2di) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtusepi64_epi8 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb128_mask ((__v2di) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtusepi64_epi8 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb256_mask ((__v4di) __A,
          (__v16qi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovusqb256mem_mask ((__v16qi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb256_mask ((__v4di) __A,
          (__v16qi) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtusepi64_epi8 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqb256_mask ((__v4di) __A,
          (__v16qi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi64_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A,
        (__v8hi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovqw128mem_mask ((__v8hi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A,
        (__v8hi)__O,
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi64_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqw128_mask ((__v2di) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi64_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A,
        (__v8hi)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A,
        (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi64_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqw256_mask ((__v4di) __A,
        (__v8hi)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsepi64_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw128_mask ((__v2di) __A,
         (__v8hi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovsqw128mem_mask ((__v8hi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw128_mask ((__v2di) __A,
         (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtsepi64_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw128_mask ((__v2di) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsepi64_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw256_mask ((__v4di) __A,
         (__v8hi)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovsqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw256_mask ((__v4di) __A,
         (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtsepi64_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqw256_mask ((__v4di) __A,
         (__v8hi)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtusepi64_epi16 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw128_mask ((__v2di) __A,
          (__v8hi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovusqw128mem_mask ((__v8hi *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_epi16 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw128_mask ((__v2di) __A,
          (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtusepi64_epi16 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw128_mask ((__v2di) __A,
          (__v8hi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtusepi64_epi16 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw256_mask ((__v4di) __A,
          (__v8hi)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovusqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw256_mask ((__v4di) __A,
          (__v8hi) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtusepi64_epi16 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqw256_mask ((__v4di) __A,
          (__v8hi)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtepi64_epi32 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A,
        (__v4si)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovqd128mem_mask ((__v4si *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A,
        (__v4si) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi64_epi32 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovqd128_mask ((__v2di) __A,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtepi64_epi32 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
        (__v4si)
        _mm_undefined_si128 (),
        (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovqd256mem_mask ((__v4si *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
        (__v4si) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi64_epi32 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovqd256_mask ((__v4di) __A,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsepi64_epi32 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd128_mask ((__v2di) __A,
         (__v4si)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovsqd128mem_mask ((__v4si *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtsepi64_epi32 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd128_mask ((__v2di) __A,
         (__v4si) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtsepi64_epi32 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd128_mask ((__v2di) __A,
         (__v4si)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtsepi64_epi32 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd256_mask ((__v4di) __A,
         (__v4si)
         _mm_undefined_si128 (),
         (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_storeu_epi32 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovsqd256mem_mask ((__v4si *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtsepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd256_mask ((__v4di) __A,
         (__v4si)__O,
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtsepi64_epi32 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovsqd256_mask ((__v4di) __A,
         (__v4si)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtusepi64_epi32 (__m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd128_mask ((__v2di) __A,
          (__v4si)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A)
{
  __builtin_ia32_pmovusqd128mem_mask ((__v4si *) __P, (__v2di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtusepi64_epi32 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd128_mask ((__v2di) __A,
          (__v4si) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtusepi64_epi32 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd128_mask ((__v2di) __A,
          (__v4si)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtusepi64_epi32 (__m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd256_mask ((__v4di) __A,
          (__v4si)
          _mm_undefined_si128 (),
          (__mmask8) -1);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_storeu_epi32 (void * __P, __mmask8 __M, __m256i __A)
{
  __builtin_ia32_pmovusqd256mem_mask ((__v4si *) __P, (__v4di) __A, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtusepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd256_mask ((__v4di) __A,
          (__v4si) __O, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtusepi64_epi32 (__mmask8 __M, __m256i __A)
{
  return (__m128i) __builtin_ia32_pmovusqd256_mask ((__v4di) __A,
          (__v4si)
          _mm_setzero_si128 (),
          __M);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcastss_ps (__m256 __O, __mmask8 __M, __m128 __A)
{
  return (__m256) __builtin_ia32_broadcastss256_mask ((__v4sf) __A,
            (__v8sf) __O,
            __M);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcastss_ps (__mmask8 __M, __m128 __A)
{
  return (__m256) __builtin_ia32_broadcastss256_mask ((__v4sf) __A,
            (__v8sf)
            _mm256_setzero_ps (),
            __M);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_broadcastss_ps (__m128 __O, __mmask8 __M, __m128 __A)
{
  return (__m128) __builtin_ia32_broadcastss128_mask ((__v4sf) __A,
            (__v4sf) __O,
            __M);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_broadcastss_ps (__mmask8 __M, __m128 __A)
{
  return (__m128) __builtin_ia32_broadcastss128_mask ((__v4sf) __A,
            (__v4sf)
            _mm_setzero_ps (),
            __M);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcastsd_pd (__m256d __O, __mmask8 __M, __m128d __A)
{
  return (__m256d) __builtin_ia32_broadcastsd256_mask ((__v2df) __A,
             (__v4df) __O,
             __M);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcastsd_pd (__mmask8 __M, __m128d __A)
{
  return (__m256d) __builtin_ia32_broadcastsd256_mask ((__v2df) __A,
             (__v4df)
             _mm256_setzero_pd (),
             __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcastd_epi32 (__m256i __O, __mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_pbroadcastd256_mask ((__v4si) __A,
             (__v8si) __O,
             __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcastd_epi32 (__mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_pbroadcastd256_mask ((__v4si) __A,
             (__v8si)
             _mm256_setzero_si256 (),
             __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_set1_epi32 (__m256i __O, __mmask8 __M, int __A)
{
  return (__m256i) __builtin_ia32_pbroadcastd256_gpr_mask (__A, (__v8si) __O,
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_set1_epi32 (__mmask8 __M, int __A)
{
  return (__m256i) __builtin_ia32_pbroadcastd256_gpr_mask (__A,
          (__v8si)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_broadcastd_epi32 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pbroadcastd128_mask ((__v4si) __A,
             (__v4si) __O,
             __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_broadcastd_epi32 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pbroadcastd128_mask ((__v4si) __A,
             (__v4si)
             _mm_setzero_si128 (),
             __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_set1_epi32 (__m128i __O, __mmask8 __M, int __A)
{
  return (__m128i) __builtin_ia32_pbroadcastd128_gpr_mask (__A, (__v4si) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_set1_epi32 (__mmask8 __M, int __A)
{
  return (__m128i)
  __builtin_ia32_pbroadcastd128_gpr_mask (__A,
       (__v4si) _mm_setzero_si128 (),
       __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcastq_epi64 (__m256i __O, __mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_pbroadcastq256_mask ((__v2di) __A,
             (__v4di) __O,
             __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcastq_epi64 (__mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_pbroadcastq256_mask ((__v2di) __A,
             (__v4di)
             _mm256_setzero_si256 (),
             __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_set1_epi64 (__m256i __O, __mmask8 __M, long long __A)
{
  return (__m256i) __builtin_ia32_pbroadcastq256_gpr_mask (__A, (__v4di) __O,
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_set1_epi64 (__mmask8 __M, long long __A)
{
  return (__m256i) __builtin_ia32_pbroadcastq256_gpr_mask (__A,
          (__v4di)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_broadcastq_epi64 (__m128i __O, __mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pbroadcastq128_mask ((__v2di) __A,
             (__v2di) __O,
             __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_broadcastq_epi64 (__mmask8 __M, __m128i __A)
{
  return (__m128i) __builtin_ia32_pbroadcastq128_mask ((__v2di) __A,
             (__v2di)
             _mm_setzero_si128 (),
             __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_set1_epi64 (__m128i __O, __mmask8 __M, long long __A)
{
  return (__m128i) __builtin_ia32_pbroadcastq128_gpr_mask (__A, (__v2di) __O,
          __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_set1_epi64 (__mmask8 __M, long long __A)
{
  return (__m128i)
  __builtin_ia32_pbroadcastq128_gpr_mask (__A,
       (__v2di) _mm_setzero_si128 (),
       __M);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_f32x4 (__m128 __A)
{
  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
                (__v8sf)_mm256_undefined_pd (),
         (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcast_f32x4 (__m256 __O, __mmask8 __M, __m128 __A)
{
  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
         (__v8sf) __O,
         __M);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcast_f32x4 (__mmask8 __M, __m128 __A)
{
  return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcast_i32x4 (__m128i __A)
{
  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
          __A,
                 (__v8si)_mm256_undefined_si256 (),
          (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_broadcast_i32x4 (__m256i __O, __mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
          __A,
          (__v8si)
          __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_broadcast_i32x4 (__mmask8 __M, __m128i __A)
{
  return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
          __A,
          (__v8si)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi8_epi32 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxbd256_mask ((__v16qi) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi8_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxbd256_mask ((__v16qi) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi8_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxbd128_mask ((__v16qi) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi8_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxbd128_mask ((__v16qi) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi8_epi64 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxbq256_mask ((__v16qi) __A,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxbq256_mask ((__v16qi) __A,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi8_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxbq128_mask ((__v16qi) __A,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxbq128_mask ((__v16qi) __A,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi16_epi32 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxwd256_mask ((__v8hi) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi16_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxwd256_mask ((__v8hi) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi16_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxwd128_mask ((__v8hi) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi16_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxwd128_mask ((__v8hi) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi16_epi64 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxwq256_mask ((__v8hi) __A,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovsxwq256_mask ((__v8hi) __A,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi16_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxwq128_mask ((__v8hi) __A,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovsxwq128_mask ((__v8hi) __A,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepi32_epi64 (__m256i __W, __mmask8 __U, __m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxdq256_mask ((__v4si) __X,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepi32_epi64 (__mmask8 __U, __m128i __X)
{
  return (__m256i) __builtin_ia32_pmovsxdq256_mask ((__v4si) __X,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepi32_epi64 (__m128i __W, __mmask8 __U, __m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxdq128_mask ((__v4si) __X,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepi32_epi64 (__mmask8 __U, __m128i __X)
{
  return (__m128i) __builtin_ia32_pmovsxdq128_mask ((__v4si) __X,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu8_epi32 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxbd256_mask ((__v16qi) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu8_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxbd256_mask ((__v16qi) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu8_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxbd128_mask ((__v16qi) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu8_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxbd128_mask ((__v16qi) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu8_epi64 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxbq256_mask ((__v16qi) __A,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxbq256_mask ((__v16qi) __A,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu8_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxbq128_mask ((__v16qi) __A,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu8_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxbq128_mask ((__v16qi) __A,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu16_epi32 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxwd256_mask ((__v8hi) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu16_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxwd256_mask ((__v8hi) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu16_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxwd128_mask ((__v8hi) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu16_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxwd128_mask ((__v8hi) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu16_epi64 (__m256i __W, __mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxwq256_mask ((__v8hi) __A,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m256i) __builtin_ia32_pmovzxwq256_mask ((__v8hi) __A,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu16_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxwq128_mask ((__v8hi) __A,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu16_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_pmovzxwq128_mask ((__v8hi) __A,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtepu32_epi64 (__m256i __W, __mmask8 __U, __m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxdq256_mask ((__v4si) __X,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtepu32_epi64 (__mmask8 __U, __m128i __X)
{
  return (__m256i) __builtin_ia32_pmovzxdq256_mask ((__v4si) __X,
          (__v4di)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtepu32_epi64 (__m128i __W, __mmask8 __U, __m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxdq128_mask ((__v4si) __X,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtepu32_epi64 (__mmask8 __U, __m128i __X)
{
  return (__m128i) __builtin_ia32_pmovzxdq128_mask ((__v4si) __X,
          (__v2di)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rcp14_pd (__m256d __A)
{
  return (__m256d) __builtin_ia32_rcp14pd256_mask ((__v4df) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rcp14_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_rcp14pd256_mask ((__v4df) __A,
           (__v4df) __W,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rcp14_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_rcp14pd256_mask ((__v4df) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp14_pd (__m128d __A)
{
  return (__m128d) __builtin_ia32_rcp14pd128_mask ((__v2df) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rcp14_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_rcp14pd128_mask ((__v2df) __A,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rcp14_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_rcp14pd128_mask ((__v2df) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rcp14_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_rcp14ps256_mask ((__v8sf) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rcp14_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_rcp14ps256_mask ((__v8sf) __A,
          (__v8sf) __W,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rcp14_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_rcp14ps256_mask ((__v8sf) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rcp14_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_rcp14ps128_mask ((__v4sf) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rcp14_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_rcp14ps128_mask ((__v4sf) __A,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rcp14_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_rcp14ps128_mask ((__v4sf) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rsqrt14_pd (__m256d __A)
{
  return (__m256d) __builtin_ia32_rsqrt14pd256_mask ((__v4df) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rsqrt14_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_rsqrt14pd256_mask ((__v4df) __A,
           (__v4df) __W,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rsqrt14_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_rsqrt14pd256_mask ((__v4df) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt14_pd (__m128d __A)
{
  return (__m128d) __builtin_ia32_rsqrt14pd128_mask ((__v2df) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rsqrt14_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_rsqrt14pd128_mask ((__v2df) __A,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rsqrt14_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_rsqrt14pd128_mask ((__v2df) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rsqrt14_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_rsqrt14ps256_mask ((__v8sf) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rsqrt14_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_rsqrt14ps256_mask ((__v8sf) __A,
          (__v8sf) __W,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rsqrt14_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_rsqrt14ps256_mask ((__v8sf) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rsqrt14_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_rsqrt14ps128_mask ((__v4sf) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rsqrt14_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_rsqrt14ps128_mask ((__v4sf) __A,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rsqrt14_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_rsqrt14ps128_mask ((__v4sf) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sqrt_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_sqrtpd256_mask ((__v4df) __A,
        (__v4df) __W,
        (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sqrt_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_sqrtpd256_mask ((__v4df) __A,
        (__v4df)
        _mm256_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sqrt_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_sqrtpd128_mask ((__v2df) __A,
        (__v2df) __W,
        (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sqrt_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_sqrtpd128_mask ((__v2df) __A,
        (__v2df)
        _mm_setzero_pd (),
        (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sqrt_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_sqrtps256_mask ((__v8sf) __A,
       (__v8sf) __W,
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sqrt_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_sqrtps256_mask ((__v8sf) __A,
       (__v8sf)
       _mm256_setzero_ps (),
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sqrt_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_sqrtps128_mask ((__v4sf) __A,
       (__v4sf) __W,
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sqrt_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_sqrtps128_mask ((__v4sf) __A,
       (__v4sf)
       _mm_setzero_ps (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_add_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_paddd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_add_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_paddd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_add_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_paddq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_add_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_paddq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sub_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_psubd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sub_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_psubd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sub_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_psubq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sub_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_psubq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_add_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_paddd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_add_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_paddd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_add_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_paddq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_add_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_paddq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sub_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psubd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sub_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psubd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sub_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psubq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sub_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psubq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_getexp_ps (__m256 __A)
{
  return (__m256) __builtin_ia32_getexpps256_mask ((__v8sf) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_getexp_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_getexpps256_mask ((__v8sf) __A,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_getexp_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_getexpps256_mask ((__v8sf) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_getexp_pd (__m256d __A)
{
  return (__m256d) __builtin_ia32_getexppd256_mask ((__v4df) __A,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_getexp_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_getexppd256_mask ((__v4df) __A,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_getexp_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_getexppd256_mask ((__v4df) __A,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_getexp_ps (__m128 __A)
{
  return (__m128) __builtin_ia32_getexpps128_mask ((__v4sf) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_getexp_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_getexpps128_mask ((__v4sf) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_getexp_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_getexpps128_mask ((__v4sf) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_getexp_pd (__m128d __A)
{
  return (__m128d) __builtin_ia32_getexppd128_mask ((__v2df) __A,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_getexp_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_getexppd128_mask ((__v2df) __A,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_getexp_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_getexppd128_mask ((__v2df) __A,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srl_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_psrld256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srl_epi32 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psrld256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srl_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psrld128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srl_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psrld128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srl_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_psrlq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srl_epi64 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psrlq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srl_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psrlq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srl_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psrlq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_and_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pandd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_and_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pandd256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_scalef_pd (__m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_scalefpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_scalef_pd (__m256d __W, __mmask8 __U, __m256d __A,
         __m256d __B)
{
  return (__m256d) __builtin_ia32_scalefpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_scalef_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_scalefpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_scalef_ps (__m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_scalefps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_scalef_ps (__m256 __W, __mmask8 __U, __m256 __A,
         __m256 __B)
{
  return (__m256) __builtin_ia32_scalefps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_scalef_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_scalefps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_scalef_pd (__m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_scalefpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_scalef_pd (__m128d __W, __mmask8 __U, __m128d __A,
      __m128d __B)
{
  return (__m128d) __builtin_ia32_scalefpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_scalef_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_scalefpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_scalef_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_scalefps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_scalef_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_scalefps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_scalef_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_scalefps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmadd_pd (__m256d __A, __mmask8 __U, __m256d __B,
        __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df) __C,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmadd_pd (__m256d __A, __m256d __B, __m256d __C,
         __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_mask3 ((__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmadd_pd (__mmask8 __U, __m256d __A, __m256d __B,
         __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_maskz ((__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmadd_pd (__m128d __A, __mmask8 __U, __m128d __B, __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df) __C,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmadd_pd (__m128d __A, __m128d __B, __m128d __C,
      __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_mask3 ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmadd_pd (__mmask8 __U, __m128d __A, __m128d __B,
      __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_maskz ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmadd_ps (__m256 __A, __mmask8 __U, __m256 __B, __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf) __C,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmadd_ps (__m256 __A, __m256 __B, __m256 __C,
         __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfmaddps256_mask3 ((__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmadd_ps (__mmask8 __U, __m256 __A, __m256 __B,
         __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_maskz ((__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmadd_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf) __C,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmadd_ps (__m128 __A, __m128 __B, __m128 __C, __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfmaddps128_mask3 ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmadd_ps (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_maskz ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmsub_pd (__m256d __A, __mmask8 __U, __m256d __B,
        __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_mask ((__v4df) __A,
          (__v4df) __B,
          -(__v4df) __C,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmsub_pd (__m256d __A, __m256d __B, __m256d __C,
         __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfmsubpd256_mask3 ((__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmsub_pd (__mmask8 __U, __m256d __A, __m256d __B,
         __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_maskz ((__v4df) __A,
           (__v4df) __B,
           -(__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmsub_pd (__m128d __A, __mmask8 __U, __m128d __B, __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_mask ((__v2df) __A,
          (__v2df) __B,
          -(__v2df) __C,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmsub_pd (__m128d __A, __m128d __B, __m128d __C,
      __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfmsubpd128_mask3 ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmsub_pd (__mmask8 __U, __m128d __A, __m128d __B,
      __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_maskz ((__v2df) __A,
           (__v2df) __B,
           -(__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmsub_ps (__m256 __A, __mmask8 __U, __m256 __B, __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         -(__v8sf) __C,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmsub_ps (__m256 __A, __m256 __B, __m256 __C,
         __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfmsubps256_mask3 ((__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmsub_ps (__mmask8 __U, __m256 __A, __m256 __B,
         __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_maskz ((__v8sf) __A,
          (__v8sf) __B,
          -(__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmsub_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         -(__v4sf) __C,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmsub_ps (__m128 __A, __m128 __B, __m128 __C, __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfmsubps128_mask3 ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmsub_ps (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_maskz ((__v4sf) __A,
          (__v4sf) __B,
          -(__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmaddsub_pd (__m256d __A, __mmask8 __U, __m256d __B,
    __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddsubpd256_mask ((__v4df) __A,
             (__v4df) __B,
             (__v4df) __C,
             (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmaddsub_pd (__m256d __A, __m256d __B, __m256d __C,
     __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfmaddsubpd256_mask3 ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __C,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmaddsub_pd (__mmask8 __U, __m256d __A, __m256d __B,
     __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddsubpd256_maskz ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __C,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmaddsub_pd (__m128d __A, __mmask8 __U, __m128d __B,
        __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddsubpd128_mask ((__v2df) __A,
             (__v2df) __B,
             (__v2df) __C,
             (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmaddsub_pd (__m128d __A, __m128d __B, __m128d __C,
         __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfmaddsubpd128_mask3 ((__v2df) __A,
       (__v2df) __B,
       (__v2df) __C,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmaddsub_pd (__mmask8 __U, __m128d __A, __m128d __B,
         __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddsubpd128_maskz ((__v2df) __A,
       (__v2df) __B,
       (__v2df) __C,
       (__mmask8)
       __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmaddsub_ps (__m256 __A, __mmask8 __U, __m256 __B,
    __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddsubps256_mask ((__v8sf) __A,
            (__v8sf) __B,
            (__v8sf) __C,
            (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmaddsub_ps (__m256 __A, __m256 __B, __m256 __C,
     __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfmaddsubps256_mask3 ((__v8sf) __A,
             (__v8sf) __B,
             (__v8sf) __C,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmaddsub_ps (__mmask8 __U, __m256 __A, __m256 __B,
     __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddsubps256_maskz ((__v8sf) __A,
             (__v8sf) __B,
             (__v8sf) __C,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmaddsub_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddsubps128_mask ((__v4sf) __A,
            (__v4sf) __B,
            (__v4sf) __C,
            (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmaddsub_ps (__m128 __A, __m128 __B, __m128 __C,
         __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfmaddsubps128_mask3 ((__v4sf) __A,
             (__v4sf) __B,
             (__v4sf) __C,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmaddsub_ps (__mmask8 __U, __m128 __A, __m128 __B,
         __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddsubps128_maskz ((__v4sf) __A,
             (__v4sf) __B,
             (__v4sf) __C,
             (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmsubadd_pd (__m256d __A, __mmask8 __U, __m256d __B,
    __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddsubpd256_mask ((__v4df) __A,
             (__v4df) __B,
             -(__v4df) __C,
             (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmsubadd_pd (__m256d __A, __m256d __B, __m256d __C,
     __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfmsubaddpd256_mask3 ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __C,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmsubadd_pd (__mmask8 __U, __m256d __A, __m256d __B,
     __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddsubpd256_maskz ((__v4df) __A,
       (__v4df) __B,
       -(__v4df) __C,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmsubadd_pd (__m128d __A, __mmask8 __U, __m128d __B,
        __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddsubpd128_mask ((__v2df) __A,
             (__v2df) __B,
             -(__v2df) __C,
             (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmsubadd_pd (__m128d __A, __m128d __B, __m128d __C,
         __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfmsubaddpd128_mask3 ((__v2df) __A,
       (__v2df) __B,
       (__v2df) __C,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmsubadd_pd (__mmask8 __U, __m128d __A, __m128d __B,
         __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddsubpd128_maskz ((__v2df) __A,
       (__v2df) __B,
       -(__v2df) __C,
       (__mmask8)
       __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fmsubadd_ps (__m256 __A, __mmask8 __U, __m256 __B,
    __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddsubps256_mask ((__v8sf) __A,
            (__v8sf) __B,
            -(__v8sf) __C,
            (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fmsubadd_ps (__m256 __A, __m256 __B, __m256 __C,
     __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfmsubaddps256_mask3 ((__v8sf) __A,
             (__v8sf) __B,
             (__v8sf) __C,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fmsubadd_ps (__mmask8 __U, __m256 __A, __m256 __B,
     __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddsubps256_maskz ((__v8sf) __A,
             (__v8sf) __B,
             -(__v8sf) __C,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fmsubadd_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddsubps128_mask ((__v4sf) __A,
            (__v4sf) __B,
            -(__v4sf) __C,
            (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fmsubadd_ps (__m128 __A, __m128 __B, __m128 __C,
         __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfmsubaddps128_mask3 ((__v4sf) __A,
             (__v4sf) __B,
             (__v4sf) __C,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fmsubadd_ps (__mmask8 __U, __m128 __A, __m128 __B,
         __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddsubps128_maskz ((__v4sf) __A,
             (__v4sf) __B,
             -(__v4sf) __C,
             (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fnmadd_pd (__m256d __A, __mmask8 __U, __m256d __B,
         __m256d __C)
{
  return (__m256d) __builtin_ia32_vfnmaddpd256_mask ((__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fnmadd_pd (__m256d __A, __m256d __B, __m256d __C,
   __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_mask3 (-(__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fnmadd_pd (__mmask8 __U, __m256d __A, __m256d __B,
   __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_maskz (-(__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fnmadd_pd (__m128d __A, __mmask8 __U, __m128d __B,
      __m128d __C)
{
  return (__m128d) __builtin_ia32_vfnmaddpd128_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fnmadd_pd (__m128d __A, __m128d __B, __m128d __C,
       __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_mask3 (-(__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fnmadd_pd (__mmask8 __U, __m128d __A, __m128d __B,
       __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_maskz (-(__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fnmadd_ps (__m256 __A, __mmask8 __U, __m256 __B,
         __m256 __C)
{
  return (__m256) __builtin_ia32_vfnmaddps256_mask ((__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fnmadd_ps (__m256 __A, __m256 __B, __m256 __C,
   __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfmaddps256_mask3 (-(__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fnmadd_ps (__mmask8 __U, __m256 __A, __m256 __B,
   __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_maskz (-(__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fnmadd_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfnmaddps128_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fnmadd_ps (__m128 __A, __m128 __B, __m128 __C, __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfmaddps128_mask3 (-(__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fnmadd_ps (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_maskz (-(__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fnmsub_pd (__m256d __A, __mmask8 __U, __m256d __B,
         __m256d __C)
{
  return (__m256d) __builtin_ia32_vfnmsubpd256_mask ((__v4df) __A,
           (__v4df) __B,
           (__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fnmsub_pd (__m256d __A, __m256d __B, __m256d __C,
   __mmask8 __U)
{
  return (__m256d) __builtin_ia32_vfnmsubpd256_mask3 ((__v4df) __A,
            (__v4df) __B,
            (__v4df) __C,
            (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fnmsub_pd (__mmask8 __U, __m256d __A, __m256d __B,
   __m256d __C)
{
  return (__m256d) __builtin_ia32_vfmaddpd256_maskz (-(__v4df) __A,
           (__v4df) __B,
           -(__v4df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fnmsub_pd (__m128d __A, __mmask8 __U, __m128d __B,
      __m128d __C)
{
  return (__m128d) __builtin_ia32_vfnmsubpd128_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fnmsub_pd (__m128d __A, __m128d __B, __m128d __C,
       __mmask8 __U)
{
  return (__m128d) __builtin_ia32_vfnmsubpd128_mask3 ((__v2df) __A,
            (__v2df) __B,
            (__v2df) __C,
            (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fnmsub_pd (__mmask8 __U, __m128d __A, __m128d __B,
       __m128d __C)
{
  return (__m128d) __builtin_ia32_vfmaddpd128_maskz (-(__v2df) __A,
           (__v2df) __B,
           -(__v2df) __C,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_fnmsub_ps (__m256 __A, __mmask8 __U, __m256 __B,
         __m256 __C)
{
  return (__m256) __builtin_ia32_vfnmsubps256_mask ((__v8sf) __A,
          (__v8sf) __B,
          (__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask3_fnmsub_ps (__m256 __A, __m256 __B, __m256 __C,
   __mmask8 __U)
{
  return (__m256) __builtin_ia32_vfnmsubps256_mask3 ((__v8sf) __A,
           (__v8sf) __B,
           (__v8sf) __C,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_fnmsub_ps (__mmask8 __U, __m256 __A, __m256 __B,
   __m256 __C)
{
  return (__m256) __builtin_ia32_vfmaddps256_maskz (-(__v8sf) __A,
          (__v8sf) __B,
          -(__v8sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_fnmsub_ps (__m128 __A, __mmask8 __U, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfnmsubps128_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask3_fnmsub_ps (__m128 __A, __m128 __B, __m128 __C, __mmask8 __U)
{
  return (__m128) __builtin_ia32_vfnmsubps128_mask3 ((__v4sf) __A,
           (__v4sf) __B,
           (__v4sf) __C,
           (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_fnmsub_ps (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
{
  return (__m128) __builtin_ia32_vfmaddps128_maskz (-(__v4sf) __A,
          (__v4sf) __B,
          -(__v4sf) __C,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_and_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pandd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_and_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pandd128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_andnot_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
     __m256i __B)
{
  return (__m256i) __builtin_ia32_pandnd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_andnot_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pandnd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_andnot_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_pandnd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_andnot_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pandnd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_or_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
        __m256i __B)
{
  return (__m256i) __builtin_ia32_pord256_mask ((__v8si) __A,
      (__v8si) __B,
      (__v8si) __W,
      (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_or_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pord256_mask ((__v8si) __A,
      (__v8si) __B,
      (__v8si)
      _mm256_setzero_si256 (),
      (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_or_epi32 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pord128_mask ((__v4si) __A,
      (__v4si) __B,
      (__v4si) __W,
      (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_or_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pord128_mask ((__v4si) __A,
      (__v4si) __B,
      (__v4si)
      _mm_setzero_si128 (),
      (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_xor_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pxord256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_xor_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pxord256_mask ((__v8si) __A,
       (__v8si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_xor_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pxord128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_xor_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pxord128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtpd_ps (__m128 __W, __mmask8 __U, __m128d __A)
{
  return (__m128) __builtin_ia32_cvtpd2ps_mask ((__v2df) __A,
      (__v4sf) __W,
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtpd_ps (__mmask8 __U, __m128d __A)
{
  return (__m128) __builtin_ia32_cvtpd2ps_mask ((__v2df) __A,
      (__v4sf)
      _mm_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtpd_ps (__m128 __W, __mmask8 __U, __m256d __A)
{
  return (__m128) __builtin_ia32_cvtpd2ps256_mask ((__v4df) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtpd_ps (__mmask8 __U, __m256d __A)
{
  return (__m128) __builtin_ia32_cvtpd2ps256_mask ((__v4df) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtps_epi32 (__m256i __W, __mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvtps2dq256_mask ((__v8sf) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtps_epi32 (__mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvtps2dq256_mask ((__v8sf) __A,
          (__v8si)
          _mm256_setzero_si256 (),
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtps_epi32 (__m128i __W, __mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvtps2dq128_mask ((__v4sf) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtps_epi32 (__mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvtps2dq128_mask ((__v4sf) __A,
          (__v4si)
          _mm_setzero_si128 (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cvtps_epu32 (__m256 __A)
{
  return (__m256i) __builtin_ia32_cvtps2udq256_mask ((__v8sf) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtps_epu32 (__m256i __W, __mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvtps2udq256_mask ((__v8sf) __A,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtps_epu32 (__mmask8 __U, __m256 __A)
{
  return (__m256i) __builtin_ia32_cvtps2udq256_mask ((__v8sf) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_epu32 (__m128 __A)
{
  return (__m128i) __builtin_ia32_cvtps2udq128_mask ((__v4sf) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtps_epu32 (__m128i __W, __mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvtps2udq128_mask ((__v4sf) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtps_epu32 (__mmask8 __U, __m128 __A)
{
  return (__m128i) __builtin_ia32_cvtps2udq128_mask ((__v4sf) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_movedup_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_movddup256_mask ((__v4df) __A,
         (__v4df) __W,
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_movedup_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_movddup256_mask ((__v4df) __A,
         (__v4df)
         _mm256_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_movedup_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_movddup128_mask ((__v2df) __A,
         (__v2df) __W,
         (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_movedup_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_movddup128_mask ((__v2df) __A,
         (__v2df)
         _mm_setzero_pd (),
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_movehdup_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_movehdup_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_movehdup_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_movehdup_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_moveldup_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_moveldup_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_moveldup_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_moveldup_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpackhi_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
    __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckhdq128_mask ((__v4si) __A,
           (__v4si) __B,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpackhi_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckhdq128_mask ((__v4si) __A,
           (__v4si) __B,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpackhi_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
       __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckhdq256_mask ((__v8si) __A,
           (__v8si) __B,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpackhi_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckhdq256_mask ((__v8si) __A,
           (__v8si) __B,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpackhi_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
    __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckhqdq128_mask ((__v2di) __A,
            (__v2di) __B,
            (__v2di) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpackhi_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckhqdq128_mask ((__v2di) __A,
            (__v2di) __B,
            (__v2di)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpackhi_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
       __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckhqdq256_mask ((__v4di) __A,
            (__v4di) __B,
            (__v4di) __W,
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpackhi_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckhqdq256_mask ((__v4di) __A,
            (__v4di) __B,
            (__v4di)
            _mm256_setzero_si256 (),
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpacklo_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
    __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckldq128_mask ((__v4si) __A,
           (__v4si) __B,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpacklo_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_punpckldq128_mask ((__v4si) __A,
           (__v4si) __B,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpacklo_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
       __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckldq256_mask ((__v8si) __A,
           (__v8si) __B,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpacklo_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_punpckldq256_mask ((__v8si) __A,
           (__v8si) __B,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpacklo_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
    __m128i __B)
{
  return (__m128i) __builtin_ia32_punpcklqdq128_mask ((__v2di) __A,
            (__v2di) __B,
            (__v2di) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpacklo_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_punpcklqdq128_mask ((__v2di) __A,
            (__v2di) __B,
            (__v2di)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpacklo_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
       __m256i __B)
{
  return (__m256i) __builtin_ia32_punpcklqdq256_mask ((__v4di) __A,
            (__v4di) __B,
            (__v4di) __W,
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpacklo_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_punpcklqdq256_mask ((__v4di) __A,
            (__v4di) __B,
            (__v4di)
            _mm256_setzero_si256 (),
            (__mmask8) __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epu32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __A,
         (__v4si) __B, 0,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqd128_mask ((__v4si) __A,
          (__v4si) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpeq_epu32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __A,
         (__v4si) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpeq_epi32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqd128_mask ((__v4si) __A,
          (__v4si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epu32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __A,
         (__v8si) __B, 0,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqd256_mask ((__v8si) __A,
          (__v8si) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpeq_epu32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __A,
         (__v8si) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpeq_epi32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqd256_mask ((__v8si) __A,
          (__v8si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epu64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __A,
         (__v2di) __B, 0,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpeq_epi64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq128_mask ((__v2di) __A,
          (__v2di) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpeq_epu64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __A,
         (__v2di) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpeq_epi64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq128_mask ((__v2di) __A,
          (__v2di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epu64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __A,
         (__v4di) __B, 0,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpeq_epi64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq256_mask ((__v4di) __A,
          (__v4di) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpeq_epu64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __A,
         (__v4di) __B, 0, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpeq_epi64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpeqq256_mask ((__v4di) __A,
          (__v4di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epu32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __A,
         (__v4si) __B, 6,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtd128_mask ((__v4si) __A,
          (__v4si) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpgt_epu32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __A,
         (__v4si) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpgt_epi32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtd128_mask ((__v4si) __A,
          (__v4si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epu32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __A,
         (__v8si) __B, 6,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtd256_mask ((__v8si) __A,
          (__v8si) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpgt_epu32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __A,
         (__v8si) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpgt_epi32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtd256_mask ((__v8si) __A,
          (__v8si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epu64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __A,
         (__v2di) __B, 6,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpgt_epi64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq128_mask ((__v2di) __A,
          (__v2di) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpgt_epu64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __A,
         (__v2di) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpgt_epi64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq128_mask ((__v2di) __A,
          (__v2di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epu64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __A,
         (__v4di) __B, 6,
         (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpgt_epi64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq256_mask ((__v4di) __A,
          (__v4di) __B,
          (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpgt_epu64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __A,
         (__v4di) __B, 6, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpgt_epi64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_pcmpgtq256_mask ((__v4di) __A,
          (__v4di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_test_epi32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestmd128 ((__v4si) __A,
            (__v4si) __B,
            (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_test_epi32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestmd128 ((__v4si) __A,
            (__v4si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_test_epi32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestmd256 ((__v8si) __A,
            (__v8si) __B,
            (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_test_epi32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestmd256 ((__v8si) __A,
            (__v8si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_test_epi64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq128 ((__v2di) __A,
            (__v2di) __B,
            (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_test_epi64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq128 ((__v2di) __A,
            (__v2di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_test_epi64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq256 ((__v4di) __A,
            (__v4di) __B,
            (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_test_epi64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestmq256 ((__v4di) __A,
            (__v4di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_testn_epi32_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmd128 ((__v4si) __A,
      (__v4si) __B,
      (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_testn_epi32_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmd128 ((__v4si) __A,
      (__v4si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testn_epi32_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmd256 ((__v8si) __A,
      (__v8si) __B,
      (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_testn_epi32_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmd256 ((__v8si) __A,
      (__v8si) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_testn_epi64_mask (__m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq128 ((__v2di) __A,
      (__v2di) __B,
      (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_testn_epi64_mask (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq128 ((__v2di) __A,
      (__v2di) __B, __U);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_testn_epi64_mask (__m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq256 ((__v4di) __A,
      (__v4di) __B,
      (__mmask8) -1);
}

extern __inline __mmask8
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_testn_epi64_mask (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__mmask8) __builtin_ia32_ptestnmq256 ((__v4di) __A,
      (__v4di) __B, __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compress_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_compressdf256_mask ((__v4df) __A,
            (__v4df) __W,
            (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_compress_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_compressdf256_mask ((__v4df) __A,
            (__v4df)
            _mm256_setzero_pd (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compressstoreu_pd (void *__P, __mmask8 __U, __m256d __A)
{
  __builtin_ia32_compressstoredf256_mask ((__v4df *) __P,
       (__v4df) __A,
       (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compress_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_compressdf128_mask ((__v2df) __A,
            (__v2df) __W,
            (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_compress_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_compressdf128_mask ((__v2df) __A,
            (__v2df)
            _mm_setzero_pd (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compressstoreu_pd (void *__P, __mmask8 __U, __m128d __A)
{
  __builtin_ia32_compressstoredf128_mask ((__v2df *) __P,
       (__v2df) __A,
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compress_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_compresssf256_mask ((__v8sf) __A,
           (__v8sf) __W,
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_compress_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_compresssf256_mask ((__v8sf) __A,
           (__v8sf)
           _mm256_setzero_ps (),
           (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compressstoreu_ps (void *__P, __mmask8 __U, __m256 __A)
{
  __builtin_ia32_compressstoresf256_mask ((__v8sf *) __P,
       (__v8sf) __A,
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compress_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_compresssf128_mask ((__v4sf) __A,
           (__v4sf) __W,
           (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_compress_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_compresssf128_mask ((__v4sf) __A,
           (__v4sf)
           _mm_setzero_ps (),
           (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compressstoreu_ps (void *__P, __mmask8 __U, __m128 __A)
{
  __builtin_ia32_compressstoresf128_mask ((__v4sf *) __P,
       (__v4sf) __A,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compress_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_compressdi256_mask ((__v4di) __A,
            (__v4di) __W,
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_compress_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_compressdi256_mask ((__v4di) __A,
            (__v4di)
            _mm256_setzero_si256 (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compressstoreu_epi64 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_compressstoredi256_mask ((__v4di *) __P,
       (__v4di) __A,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compress_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_compressdi128_mask ((__v2di) __A,
            (__v2di) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_compress_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_compressdi128_mask ((__v2di) __A,
            (__v2di)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compressstoreu_epi64 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_compressstoredi128_mask ((__v2di *) __P,
       (__v2di) __A,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compress_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_compresssi256_mask ((__v8si) __A,
            (__v8si) __W,
            (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_compress_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_compresssi256_mask ((__v8si) __A,
            (__v8si)
            _mm256_setzero_si256 (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_compressstoreu_epi32 (void *__P, __mmask8 __U, __m256i __A)
{
  __builtin_ia32_compressstoresi256_mask ((__v8si *) __P,
       (__v8si) __A,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compress_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_compresssi128_mask ((__v4si) __A,
            (__v4si) __W,
            (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_compress_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_compresssi128_mask ((__v4si) __A,
            (__v4si)
            _mm_setzero_si128 (),
            (__mmask8) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_compressstoreu_epi32 (void *__P, __mmask8 __U, __m128i __A)
{
  __builtin_ia32_compressstoresi128_mask ((__v4si *) __P,
       (__v4si) __A,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expand_pd (__m256d __W, __mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_expanddf256_mask ((__v4df) __A,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expand_pd (__mmask8 __U, __m256d __A)
{
  return (__m256d) __builtin_ia32_expanddf256_maskz ((__v4df) __A,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expandloadu_pd (__m256d __W, __mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_expandloaddf256_mask ((__v4df *) __P,
       (__v4df) __W,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expandloadu_pd (__mmask8 __U, void const *__P)
{
  return (__m256d) __builtin_ia32_expandloaddf256_maskz ((__v4df *) __P,
        (__v4df)
        _mm256_setzero_pd (),
        (__mmask8)
        __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expand_pd (__m128d __W, __mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_expanddf128_mask ((__v2df) __A,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expand_pd (__mmask8 __U, __m128d __A)
{
  return (__m128d) __builtin_ia32_expanddf128_maskz ((__v2df) __A,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expandloadu_pd (__m128d __W, __mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_expandloaddf128_mask ((__v2df *) __P,
       (__v2df) __W,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expandloadu_pd (__mmask8 __U, void const *__P)
{
  return (__m128d) __builtin_ia32_expandloaddf128_maskz ((__v2df *) __P,
        (__v2df)
        _mm_setzero_pd (),
        (__mmask8)
        __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expand_ps (__m256 __W, __mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_expandsf256_mask ((__v8sf) __A,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expand_ps (__mmask8 __U, __m256 __A)
{
  return (__m256) __builtin_ia32_expandsf256_maskz ((__v8sf) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expandloadu_ps (__m256 __W, __mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_expandloadsf256_mask ((__v8sf *) __P,
             (__v8sf) __W,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expandloadu_ps (__mmask8 __U, void const *__P)
{
  return (__m256) __builtin_ia32_expandloadsf256_maskz ((__v8sf *) __P,
       (__v8sf)
       _mm256_setzero_ps (),
       (__mmask8)
       __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expand_ps (__m128 __W, __mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_expandsf128_mask ((__v4sf) __A,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expand_ps (__mmask8 __U, __m128 __A)
{
  return (__m128) __builtin_ia32_expandsf128_maskz ((__v4sf) __A,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expandloadu_ps (__m128 __W, __mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_expandloadsf128_mask ((__v4sf *) __P,
             (__v4sf) __W,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expandloadu_ps (__mmask8 __U, void const *__P)
{
  return (__m128) __builtin_ia32_expandloadsf128_maskz ((__v4sf *) __P,
       (__v4sf)
       _mm_setzero_ps (),
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expand_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_expanddi256_mask ((__v4di) __A,
          (__v4di) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expand_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_expanddi256_maskz ((__v4di) __A,
           (__v4di)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expandloadu_epi64 (__m256i __W, __mmask8 __U,
          void const *__P)
{
  return (__m256i) __builtin_ia32_expandloaddi256_mask ((__v4di *) __P,
       (__v4di) __W,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expandloadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_expandloaddi256_maskz ((__v4di *) __P,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expand_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_expanddi128_mask ((__v2di) __A,
          (__v2di) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expand_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_expanddi128_maskz ((__v2di) __A,
           (__v2di)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expandloadu_epi64 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_expandloaddi128_mask ((__v2di *) __P,
       (__v2di) __W,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expandloadu_epi64 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_expandloaddi128_maskz ((__v2di *) __P,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8)
        __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expand_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_expandsi256_mask ((__v8si) __A,
          (__v8si) __W,
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expand_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_expandsi256_maskz ((__v8si) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_expandloadu_epi32 (__m256i __W, __mmask8 __U,
          void const *__P)
{
  return (__m256i) __builtin_ia32_expandloadsi256_mask ((__v8si *) __P,
       (__v8si) __W,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_expandloadu_epi32 (__mmask8 __U, void const *__P)
{
  return (__m256i) __builtin_ia32_expandloadsi256_maskz ((__v8si *) __P,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expand_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_expandsi128_mask ((__v4si) __A,
          (__v4si) __W,
          (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expand_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_expandsi128_maskz ((__v4si) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_expandloadu_epi32 (__m128i __W, __mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_expandloadsi128_mask ((__v4si *) __P,
       (__v4si) __W,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_expandloadu_epi32 (__mmask8 __U, void const *__P)
{
  return (__m128i) __builtin_ia32_expandloadsi128_maskz ((__v4si *) __P,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8)
        __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutex2var_pd (__m256d __A, __m256i __I, __m256d __B)
{
  return (__m256d) __builtin_ia32_vpermt2varpd256_mask ((__v4di) __I
                 ,
       (__v4df) __A,
       (__v4df) __B,
       (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutex2var_pd (__m256d __A, __mmask8 __U, __m256i __I,
        __m256d __B)
{
  return (__m256d) __builtin_ia32_vpermt2varpd256_mask ((__v4di) __I
                 ,
       (__v4df) __A,
       (__v4df) __B,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask2_permutex2var_pd (__m256d __A, __m256i __I, __mmask8 __U,
         __m256d __B)
{
  return (__m256d) __builtin_ia32_vpermi2varpd256_mask ((__v4df) __A,
       (__v4di) __I
                 ,
       (__v4df) __B,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutex2var_pd (__mmask8 __U, __m256d __A, __m256i __I,
         __m256d __B)
{
  return (__m256d) __builtin_ia32_vpermt2varpd256_maskz ((__v4di) __I
                  ,
        (__v4df) __A,
        (__v4df) __B,
        (__mmask8)
        __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutex2var_ps (__m256 __A, __m256i __I, __m256 __B)
{
  return (__m256) __builtin_ia32_vpermt2varps256_mask ((__v8si) __I
                       ,
             (__v8sf) __A,
             (__v8sf) __B,
             (__mmask8) -1);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutex2var_ps (__m256 __A, __mmask8 __U, __m256i __I,
        __m256 __B)
{
  return (__m256) __builtin_ia32_vpermt2varps256_mask ((__v8si) __I
                       ,
             (__v8sf) __A,
             (__v8sf) __B,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask2_permutex2var_ps (__m256 __A, __m256i __I, __mmask8 __U,
         __m256 __B)
{
  return (__m256) __builtin_ia32_vpermi2varps256_mask ((__v8sf) __A,
             (__v8si) __I
                       ,
             (__v8sf) __B,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutex2var_ps (__mmask8 __U, __m256 __A, __m256i __I,
         __m256 __B)
{
  return (__m256) __builtin_ia32_vpermt2varps256_maskz ((__v8si) __I
                 ,
       (__v8sf) __A,
       (__v8sf) __B,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutex2var_epi64 (__m128i __A, __m128i __I, __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2varq128_mask ((__v2di) __I
                       ,
             (__v2di) __A,
             (__v2di) __B,
             (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutex2var_epi64 (__m128i __A, __mmask8 __U, __m128i __I,
        __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2varq128_mask ((__v2di) __I
                       ,
             (__v2di) __A,
             (__v2di) __B,
             (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask2_permutex2var_epi64 (__m128i __A, __m128i __I, __mmask8 __U,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermi2varq128_mask ((__v2di) __A,
             (__v2di) __I
                       ,
             (__v2di) __B,
             (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutex2var_epi64 (__mmask8 __U, __m128i __A, __m128i __I,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2varq128_maskz ((__v2di) __I
                 ,
       (__v2di) __A,
       (__v2di) __B,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutex2var_epi32 (__m128i __A, __m128i __I, __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2vard128_mask ((__v4si) __I
                       ,
             (__v4si) __A,
             (__v4si) __B,
             (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutex2var_epi32 (__m128i __A, __mmask8 __U, __m128i __I,
        __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2vard128_mask ((__v4si) __I
                       ,
             (__v4si) __A,
             (__v4si) __B,
             (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask2_permutex2var_epi32 (__m128i __A, __m128i __I, __mmask8 __U,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermi2vard128_mask ((__v4si) __A,
             (__v4si) __I
                       ,
             (__v4si) __B,
             (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutex2var_epi32 (__mmask8 __U, __m128i __A, __m128i __I,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_vpermt2vard128_maskz ((__v4si) __I
                 ,
       (__v4si) __A,
       (__v4si) __B,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutex2var_epi64 (__m256i __A, __m256i __I, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2varq256_mask ((__v4di) __I
                       ,
             (__v4di) __A,
             (__v4di) __B,
             (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutex2var_epi64 (__m256i __A, __mmask8 __U, __m256i __I,
    __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2varq256_mask ((__v4di) __I
                       ,
             (__v4di) __A,
             (__v4di) __B,
             (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask2_permutex2var_epi64 (__m256i __A, __m256i __I,
     __mmask8 __U, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermi2varq256_mask ((__v4di) __A,
             (__v4di) __I
                       ,
             (__v4di) __B,
             (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutex2var_epi64 (__mmask8 __U, __m256i __A,
     __m256i __I, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2varq256_maskz ((__v4di) __I
                 ,
       (__v4di) __A,
       (__v4di) __B,
       (__mmask8)
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutex2var_epi32 (__m256i __A, __m256i __I, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2vard256_mask ((__v8si) __I
                       ,
             (__v8si) __A,
             (__v8si) __B,
             (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutex2var_epi32 (__m256i __A, __mmask8 __U, __m256i __I,
    __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2vard256_mask ((__v8si) __I
                       ,
             (__v8si) __A,
             (__v8si) __B,
             (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask2_permutex2var_epi32 (__m256i __A, __m256i __I,
     __mmask8 __U, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermi2vard256_mask ((__v8si) __A,
             (__v8si) __I
                       ,
             (__v8si) __B,
             (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutex2var_epi32 (__mmask8 __U, __m256i __A,
     __m256i __I, __m256i __B)
{
  return (__m256i) __builtin_ia32_vpermt2vard256_maskz ((__v8si) __I
                 ,
       (__v8si) __A,
       (__v8si) __B,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutex2var_pd (__m128d __A, __m128i __I, __m128d __B)
{
  return (__m128d) __builtin_ia32_vpermt2varpd128_mask ((__v2di) __I
                 ,
       (__v2df) __A,
       (__v2df) __B,
       (__mmask8) -1);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutex2var_pd (__m128d __A, __mmask8 __U, __m128i __I,
     __m128d __B)
{
  return (__m128d) __builtin_ia32_vpermt2varpd128_mask ((__v2di) __I
                 ,
       (__v2df) __A,
       (__v2df) __B,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask2_permutex2var_pd (__m128d __A, __m128i __I, __mmask8 __U,
      __m128d __B)
{
  return (__m128d) __builtin_ia32_vpermi2varpd128_mask ((__v2df) __A,
       (__v2di) __I
                 ,
       (__v2df) __B,
       (__mmask8)
       __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutex2var_pd (__mmask8 __U, __m128d __A, __m128i __I,
      __m128d __B)
{
  return (__m128d) __builtin_ia32_vpermt2varpd128_maskz ((__v2di) __I
                  ,
        (__v2df) __A,
        (__v2df) __B,
        (__mmask8)
        __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_permutex2var_ps (__m128 __A, __m128i __I, __m128 __B)
{
  return (__m128) __builtin_ia32_vpermt2varps128_mask ((__v4si) __I
                       ,
             (__v4sf) __A,
             (__v4sf) __B,
             (__mmask8) -1);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutex2var_ps (__m128 __A, __mmask8 __U, __m128i __I,
     __m128 __B)
{
  return (__m128) __builtin_ia32_vpermt2varps128_mask ((__v4si) __I
                       ,
             (__v4sf) __A,
             (__v4sf) __B,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask2_permutex2var_ps (__m128 __A, __m128i __I, __mmask8 __U,
      __m128 __B)
{
  return (__m128) __builtin_ia32_vpermi2varps128_mask ((__v4sf) __A,
             (__v4si) __I
                       ,
             (__v4sf) __B,
             (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutex2var_ps (__mmask8 __U, __m128 __A, __m128i __I,
      __m128 __B)
{
  return (__m128) __builtin_ia32_vpermt2varps128_maskz ((__v4si) __I
                 ,
       (__v4sf) __A,
       (__v4sf) __B,
       (__mmask8)
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_srav_epi64 (__m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psravq128_mask ((__v2di) __X,
        (__v2di) __Y,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srav_epi64 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psravq128_mask ((__v2di) __X,
        (__v2di) __Y,
        (__v2di) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srav_epi64 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psravq128_mask ((__v2di) __X,
        (__v2di) __Y,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sllv_epi32 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sllv_epi32 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sllv_epi32 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sllv_epi32 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sllv_epi64 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv4di_mask ((__v4di) __X,
       (__v4di) __Y,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sllv_epi64 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psllv4di_mask ((__v4di) __X,
       (__v4di) __Y,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sllv_epi64 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv2di_mask ((__v2di) __X,
       (__v2di) __Y,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sllv_epi64 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psllv2di_mask ((__v2di) __X,
       (__v2di) __Y,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srav_epi32 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrav8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srav_epi32 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrav8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srav_epi32 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrav4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srav_epi32 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrav4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srlv_epi32 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srlv_epi32 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv8si_mask ((__v8si) __X,
       (__v8si) __Y,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srlv_epi32 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srlv_epi32 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv4si_mask ((__v4si) __X,
       (__v4si) __Y,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srlv_epi64 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv4di_mask ((__v4di) __X,
       (__v4di) __Y,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srlv_epi64 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psrlv4di_mask ((__v4di) __X,
       (__v4di) __Y,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_srlv_epi64 (__m128i __W, __mmask8 __U, __m128i __X,
       __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv2di_mask ((__v2di) __X,
       (__v2di) __Y,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_srlv_epi64 (__mmask8 __U, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_psrlv2di_mask ((__v2di) __X,
       (__v2di) __Y,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rolv_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rolv_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
   __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rolv_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rolv_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rolv_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
       __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rolv_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rorv_epi32 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rorv_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
   __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rorv_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rorv_epi32 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rorv_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
       __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rorv_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rolv_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rolv_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
   __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rolv_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prolvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rolv_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rolv_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
       __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rolv_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prolvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_rorv_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_rorv_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
   __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_rorv_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_prorvq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_rorv_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_rorv_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
       __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W,
        (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_rorv_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_prorvq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_srav_epi64 (__m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psravq256_mask ((__v4di) __X,
        (__v4di) __Y,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_srav_epi64 (__m256i __W, __mmask8 __U, __m256i __X,
   __m256i __Y)
{
  return (__m256i) __builtin_ia32_psravq256_mask ((__v4di) __X,
        (__v4di) __Y,
        (__v4di) __W,
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_srav_epi64 (__mmask8 __U, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_psravq256_mask ((__v4di) __X,
        (__v4di) __Y,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_and_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pandq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di) __W, __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_and_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pandq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di)
       _mm256_setzero_pd (),
       __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_and_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pandq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W, __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_and_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pandq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_pd (),
       __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_andnot_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
     __m256i __B)
{
  return (__m256i) __builtin_ia32_pandnq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W, __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_andnot_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pandnq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_pd (),
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_andnot_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
         __m128i __B)
{
  return (__m128i) __builtin_ia32_pandnq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W, __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_andnot_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pandnq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_pd (),
        __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_or_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
        __m256i __B)
{
  return (__m256i) __builtin_ia32_porq256_mask ((__v4di) __A,
      (__v4di) __B,
      (__v4di) __W,
      (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_or_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_porq256_mask ((__v4di) __A,
      (__v4di) __B,
      (__v4di)
      _mm256_setzero_si256 (),
      (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_or_epi64 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_porq128_mask ((__v2di) __A,
      (__v2di) __B,
      (__v2di) __W,
      (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_or_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_porq128_mask ((__v2di) __A,
      (__v2di) __B,
      (__v2di)
      _mm_setzero_si128 (),
      (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_xor_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pxorq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_xor_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pxorq256_mask ((__v4di) __A,
       (__v4di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_xor_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pxorq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_xor_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pxorq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_maxpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_maxpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_maxps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_maxps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_div_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_divps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_div_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_divps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_div_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_divpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_div_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_divpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_minpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_div_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_divpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_minpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_minps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_div_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_divpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_div_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_divps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_minps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_div_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_divps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_minps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mul_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_mulps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_minps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mul_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_mulps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_maxps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_maxps_mask ((__v4sf) __A,
          (__v4sf) __B,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_minpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_minpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_maxpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_maxpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mul_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_mulpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mul_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_mulpd_mask ((__v2df) __A,
           (__v2df) __B,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mul_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_mulps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf) __W,
      (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mul_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_mulps256_mask ((__v8sf) __A,
      (__v8sf) __B,
      (__v8sf)
      _mm256_setzero_ps (),
      (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mul_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256d __B)
{
  return (__m256d) __builtin_ia32_mulpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df) __W,
       (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mul_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_mulpd256_mask ((__v4df) __A,
       (__v4df) __B,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_epi64 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_epi64 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_epi64 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pminsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_epi64 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_epu64 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epi64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxsq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_max_epu64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_epu64 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_min_epu64 (__m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_epu64 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pminuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_epu64 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminuq256_mask ((__v4di) __A,
        (__v4di) __B,
        (__v4di)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_epi32 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxsd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_epi32 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxsd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_epi32 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminsd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_epi32 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pminsd256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_max_epu32 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxud256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_max_epu32 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pmaxud256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_min_epu32 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pminud256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_min_epu32 (__m256i __W, __mmask8 __M, __m256i __A,
         __m256i __B)
{
  return (__m256i) __builtin_ia32_pminud256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_epi64 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_epi64 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_epi64 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pminsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_epi64 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_epu64 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxsq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_epu64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_epu64 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_epu64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_epu64 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pminuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_epu64 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminuq128_mask ((__v2di) __A,
        (__v2di) __B,
        (__v2di)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_epi32 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxsd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_epi32 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxsd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_epi32 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminsd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_epi32 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pminsd128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_max_epu32 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxud128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_max_epu32 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pmaxud128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_min_epu32 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pminud128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_min_epu32 (__m128i __W, __mmask8 __M, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pminud128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W, __M);
}


#pragma GCC push_options
#pragma GCC target("avx512vl,avx512cd")



extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastmb_epi64 (__mmask8 __A)
{
  return (__m128i) __builtin_ia32_broadcastmb128 (__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastmb_epi64 (__mmask8 __A)
{
  return (__m256i) __builtin_ia32_broadcastmb256 (__A);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_broadcastmw_epi32 (__mmask16 __A)
{
  return (__m128i) __builtin_ia32_broadcastmw128 (__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_broadcastmw_epi32 (__mmask16 __A)
{
  return (__m256i) __builtin_ia32_broadcastmw256 (__A);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_lzcnt_epi32 (__m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntd_256_mask ((__v8si) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_lzcnt_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntd_256_mask ((__v8si) __A,
           (__v8si) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_lzcnt_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntd_256_mask ((__v8si) __A,
           (__v8si)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_lzcnt_epi64 (__m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntq_256_mask ((__v4di) __A,
           (__v4di)
           _mm256_setzero_si256 (),
           (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_lzcnt_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntq_256_mask ((__v4di) __A,
           (__v4di) __W,
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_lzcnt_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vplzcntq_256_mask ((__v4di) __A,
           (__v4di)
           _mm256_setzero_si256 (),
           (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_conflict_epi64 (__m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_conflict_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
        (__v4di) __W,
        (__mmask8)
        __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_conflict_epi64 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
        (__v4di)
        _mm256_setzero_si256 (),
        (__mmask8)
        __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_conflict_epi32 (__m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_conflict_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
        (__v8si) __W,
        (__mmask8)
        __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_conflict_epi32 (__mmask8 __U, __m256i __A)
{
  return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
        (__v8si)
        _mm256_setzero_si256 (),
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_lzcnt_epi32 (__m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntd_128_mask ((__v4si) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_lzcnt_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntd_128_mask ((__v4si) __A,
           (__v4si) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_lzcnt_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntd_128_mask ((__v4si) __A,
           (__v4si)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_lzcnt_epi64 (__m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntq_128_mask ((__v2di) __A,
           (__v2di)
           _mm_setzero_si128 (),
           (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_lzcnt_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntq_128_mask ((__v2di) __A,
           (__v2di) __W,
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_lzcnt_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vplzcntq_128_mask ((__v2di) __A,
           (__v2di)
           _mm_setzero_si128 (),
           (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_conflict_epi64 (__m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_conflict_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
        (__v2di) __W,
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_conflict_epi64 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
        (__v2di)
        _mm_setzero_si128 (),
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_conflict_epi32 (__m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_conflict_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
        (__v4si) __W,
        (__mmask8)
        __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_conflict_epi32 (__mmask8 __U, __m128i __A)
{
  return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
        (__v4si)
        _mm_setzero_si128 (),
        (__mmask8)
        __U);
}


#pragma GCC pop_options


extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpacklo_pd (__m256d __W, __mmask8 __U, __m256d __A,
    __m256d __B)
{
  return (__m256d) __builtin_ia32_unpcklpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpacklo_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_unpcklpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpacklo_pd (__m128d __W, __mmask8 __U, __m128d __A,
        __m128d __B)
{
  return (__m128d) __builtin_ia32_unpcklpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpacklo_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_unpcklpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpacklo_ps (__m256 __W, __mmask8 __U, __m256 __A,
    __m256 __B)
{
  return (__m256) __builtin_ia32_unpcklps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpackhi_pd (__m256d __W, __mmask8 __U, __m256d __A,
    __m256d __B)
{
  return (__m256d) __builtin_ia32_unpckhpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df) __W,
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpackhi_pd (__mmask8 __U, __m256d __A, __m256d __B)
{
  return (__m256d) __builtin_ia32_unpckhpd256_mask ((__v4df) __A,
          (__v4df) __B,
          (__v4df)
          _mm256_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpackhi_pd (__m128d __W, __mmask8 __U, __m128d __A,
        __m128d __B)
{
  return (__m128d) __builtin_ia32_unpckhpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df) __W,
          (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpackhi_pd (__mmask8 __U, __m128d __A, __m128d __B)
{
  return (__m128d) __builtin_ia32_unpckhpd128_mask ((__v2df) __A,
          (__v2df) __B,
          (__v2df)
          _mm_setzero_pd (),
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_unpackhi_ps (__m256 __W, __mmask8 __U, __m256 __A,
    __m256 __B)
{
  return (__m256) __builtin_ia32_unpckhps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf) __W,
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpackhi_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_unpckhps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpackhi_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpckhps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpackhi_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpckhps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cvtph_ps (__m128 __W, __mmask8 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_vcvtph2ps_mask ((__v8hi) __A,
       (__v4sf) __W,
       (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_cvtph_ps (__mmask8 __U, __m128i __A)
{
  return (__m128) __builtin_ia32_vcvtph2ps_mask ((__v8hi) __A,
       (__v4sf)
       _mm_setzero_ps (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_unpacklo_ps (__mmask8 __U, __m256 __A, __m256 __B)
{
  return (__m256) __builtin_ia32_unpcklps256_mask ((__v8sf) __A,
         (__v8sf) __B,
         (__v8sf)
         _mm256_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cvtph_ps (__m256 __W, __mmask8 __U, __m128i __A)
{
  return (__m256) __builtin_ia32_vcvtph2ps256_mask ((__v8hi) __A,
          (__v8sf) __W,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_cvtph_ps (__mmask8 __U, __m128i __A)
{
  return (__m256) __builtin_ia32_vcvtph2ps256_mask ((__v8hi) __A,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_unpacklo_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpcklps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf) __W,
         (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_unpacklo_ps (__mmask8 __U, __m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpcklps128_mask ((__v4sf) __A,
         (__v4sf) __B,
         (__v4sf)
         _mm_setzero_ps (),
         (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sra_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_psrad256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sra_epi32 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psrad256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sra_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psrad128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sra_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psrad128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_sra_epi64 (__m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sra_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sra_epi64 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psraq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_sra_epi64 (__m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) -1);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sra_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sra_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psraq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sll_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_pslld128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sll_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pslld128_mask ((__v4si) __A,
       (__v4si) __B,
       (__v4si)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_sll_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
      __m128i __B)
{
  return (__m128i) __builtin_ia32_psllq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di) __W,
       (__mmask8) __U);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_sll_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_psllq128_mask ((__v2di) __A,
       (__v2di) __B,
       (__v2di)
       _mm_setzero_si128 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sll_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_pslld256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sll_epi32 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_pslld256_mask ((__v8si) __A,
       (__v4si) __B,
       (__v8si)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_sll_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
         __m128i __B)
{
  return (__m256i) __builtin_ia32_psllq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di) __W,
       (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_sll_epi64 (__mmask8 __U, __m256i __A, __m128i __B)
{
  return (__m256i) __builtin_ia32_psllq256_mask ((__v4di) __A,
       (__v2di) __B,
       (__v4di)
       _mm256_setzero_si256 (),
       (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutexvar_ps (__m256 __W, __mmask8 __U, __m256i __X,
       __m256 __Y)
{
  return (__m256) __builtin_ia32_permvarsf256_mask ((__v8sf) __Y,
          (__v8si) __X,
          (__v8sf) __W,
          (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutexvar_ps (__mmask8 __U, __m256i __X, __m256 __Y)
{
  return (__m256) __builtin_ia32_permvarsf256_mask ((__v8sf) __Y,
          (__v8si) __X,
          (__v8sf)
          _mm256_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_permutexvar_pd (__m256i __X, __m256d __Y)
{
  return (__m256d) __builtin_ia32_permvardf256_mask ((__v4df) __Y,
           (__v4di) __X,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) -1);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutexvar_pd (__m256d __W, __mmask8 __U, __m256i __X,
       __m256d __Y)
{
  return (__m256d) __builtin_ia32_permvardf256_mask ((__v4df) __Y,
           (__v4di) __X,
           (__v4df) __W,
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutexvar_pd (__mmask8 __U, __m256i __X, __m256d __Y)
{
  return (__m256d) __builtin_ia32_permvardf256_mask ((__v4df) __Y,
           (__v4di) __X,
           (__v4df)
           _mm256_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutevar_pd (__m256d __W, __mmask8 __U, __m256d __A,
      __m256i __C)
{
  return (__m256d) __builtin_ia32_vpermilvarpd256_mask ((__v4df) __A,
       (__v4di) __C,
       (__v4df) __W,
       (__mmask8)
       __U);
}

extern __inline __m256d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutevar_pd (__mmask8 __U, __m256d __A, __m256i __C)
{
  return (__m256d) __builtin_ia32_vpermilvarpd256_mask ((__v4df) __A,
       (__v4di) __C,
       (__v4df)
       _mm256_setzero_pd (),
       (__mmask8)
       __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutevar_ps (__m256 __W, __mmask8 __U, __m256 __A,
      __m256i __C)
{
  return (__m256) __builtin_ia32_vpermilvarps256_mask ((__v8sf) __A,
             (__v8si) __C,
             (__v8sf) __W,
             (__mmask8) __U);
}

extern __inline __m256
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutevar_ps (__mmask8 __U, __m256 __A, __m256i __C)
{
  return (__m256) __builtin_ia32_vpermilvarps256_mask ((__v8sf) __A,
             (__v8si) __C,
             (__v8sf)
             _mm256_setzero_ps (),
             (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutevar_pd (__m128d __W, __mmask8 __U, __m128d __A,
   __m128i __C)
{
  return (__m128d) __builtin_ia32_vpermilvarpd_mask ((__v2df) __A,
           (__v2di) __C,
           (__v2df) __W,
           (__mmask8) __U);
}

extern __inline __m128d
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutevar_pd (__mmask8 __U, __m128d __A, __m128i __C)
{
  return (__m128d) __builtin_ia32_vpermilvarpd_mask ((__v2df) __A,
           (__v2di) __C,
           (__v2df)
           _mm_setzero_pd (),
           (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_permutevar_ps (__m128 __W, __mmask8 __U, __m128 __A,
   __m128i __C)
{
  return (__m128) __builtin_ia32_vpermilvarps_mask ((__v4sf) __A,
          (__v4si) __C,
          (__v4sf) __W,
          (__mmask8) __U);
}

extern __inline __m128
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_permutevar_ps (__mmask8 __U, __m128 __A, __m128i __C)
{
  return (__m128) __builtin_ia32_vpermilvarps_mask ((__v4sf) __A,
          (__v4si) __C,
          (__v4sf)
          _mm_setzero_ps (),
          (__mmask8) __U);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mullo_epi32 (__mmask8 __M, __m256i __A, __m256i __B)
{
  return (__m256i) __builtin_ia32_pmulld256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutexvar_epi64 (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_permvardi256_mask ((__v4di) __Y,
           (__v4di) __X,
           (__v4di)
           _mm256_setzero_si256 (),
           __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mullo_epi32 (__m256i __W, __mmask8 __M, __m256i __A,
    __m256i __B)
{
  return (__m256i) __builtin_ia32_pmulld256_mask ((__v8si) __A,
        (__v8si) __B,
        (__v8si) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mullo_epi32 (__mmask8 __M, __m128i __A, __m128i __B)
{
  return (__m128i) __builtin_ia32_pmulld128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mullo_epi32 (__m128i __W, __mmask16 __M, __m128i __A,
        __m128i __B)
{
  return (__m128i) __builtin_ia32_pmulld128_mask ((__v4si) __A,
        (__v4si) __B,
        (__v4si) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mul_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
         __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
        (__v8si) __Y,
        (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mul_epi32 (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
        (__v8si) __Y,
        (__v4di)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mul_epi32 (__m128i __W, __mmask8 __M, __m128i __X,
      __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
        (__v4si) __Y,
        (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mul_epi32 (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
        (__v4si) __Y,
        (__v2di)
        _mm_setzero_si128 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutexvar_epi64 (__m256i __W, __mmask8 __M, __m256i __X,
          __m256i __Y)
{
  return (__m256i) __builtin_ia32_permvardi256_mask ((__v4di) __Y,
           (__v4di) __X,
           (__v4di) __W,
           __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_mul_epu32 (__m256i __W, __mmask8 __M, __m256i __X,
         __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
         (__v8si) __Y,
         (__v4di) __W, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_permutexvar_epi32 (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
           (__v8si) __X,
           (__v8si)
           _mm256_setzero_si256 (),
           __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_maskz_mul_epu32 (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
         (__v8si) __Y,
         (__v4di)
         _mm256_setzero_si256 (),
         __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_mul_epu32 (__m128i __W, __mmask8 __M, __m128i __X,
      __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
         (__v4si) __Y,
         (__v2di) __W, __M);
}

extern __inline __m128i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskz_mul_epu32 (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
         (__v4si) __Y,
         (__v2di)
         _mm_setzero_si128 (),
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_permutexvar_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
          __m256i __Y)
{
  return (__m256i) __builtin_ia32_permvarsi256_mask ((__v8si) __Y,
           (__v8si) __X,
           (__v8si) __W,
           __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpneq_epu32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 4,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpneq_epu32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 4,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmplt_epu32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 1,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmplt_epu32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 1,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpge_epu32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 5,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpge_epu32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 5,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmple_epu32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 2,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmple_epu32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd256_mask ((__v8si) __X,
        (__v8si) __Y, 2,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpneq_epu64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 4,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpneq_epu64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 4,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmplt_epu64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 1,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmplt_epu64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 1,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpge_epu64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 5,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpge_epu64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 5,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmple_epu64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 2,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmple_epu64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq256_mask ((__v4di) __X,
        (__v4di) __Y, 2,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpneq_epi32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 4,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpneq_epi32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 4,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmplt_epi32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 1,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmplt_epi32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 1,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpge_epi32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 5,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpge_epi32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 5,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmple_epi32_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 2,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmple_epi32_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd256_mask ((__v8si) __X,
       (__v8si) __Y, 2,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpneq_epi64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 4,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpneq_epi64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 4,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmplt_epi64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 1,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmplt_epi64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 1,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmpge_epi64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 5,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmpge_epi64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 5,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_mask_cmple_epi64_mask (__mmask8 __M, __m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 2,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm256_cmple_epi64_mask (__m256i __X, __m256i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq256_mask ((__v4di) __X,
       (__v4di) __Y, 2,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpneq_epu32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 4,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_epu32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 4,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmplt_epu32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 1,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epu32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 1,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpge_epu32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 5,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_epu32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 5,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmple_epu32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 2,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_epu32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpd128_mask ((__v4si) __X,
        (__v4si) __Y, 2,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpneq_epu64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 4,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_epu64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 4,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmplt_epu64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 1,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epu64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 1,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpge_epu64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 5,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_epu64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 5,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmple_epu64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 2,
        (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_epu64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_ucmpq128_mask ((__v2di) __X,
        (__v2di) __Y, 2,
        (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpneq_epi32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 4,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_epi32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 4,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmplt_epi32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 1,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epi32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 1,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpge_epi32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 5,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_epi32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 5,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmple_epi32_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 2,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_epi32_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpd128_mask ((__v4si) __X,
       (__v4si) __Y, 2,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpneq_epi64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 4,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpneq_epi64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 4,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmplt_epi64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 1,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmplt_epi64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 1,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmpge_epi64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 5,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmpge_epi64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 5,
       (__mmask8) -1);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_mask_cmple_epi64_mask (__mmask8 __M, __m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 2,
       (__mmask8) __M);
}

extern __inline __mmask8
  __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmple_epi64_mask (__m128i __X, __m128i __Y)
{
  return (__mmask8) __builtin_ia32_cmpq128_mask ((__v2di) __X,
       (__v2di) __Y, 2,
       (__mmask8) -1);
}
# 13603 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512vlintrin.h" 3 4
#pragma GCC pop_options
# 54 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512bwintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512bwintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512bw")




typedef short __v32hi __attribute__ ((__vector_size__ (64)));
typedef char __v64qi __attribute__ ((__vector_size__ (64)));

typedef unsigned long long __mmask64;

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktest_mask32_u8 (__mmask32 __A, __mmask32 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_ktestcsi (__A, __B);
  return (unsigned char) __builtin_ia32_ktestzsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktest_mask64_u8 (__mmask64 __A, __mmask64 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_ktestcdi (__A, __B);
  return (unsigned char) __builtin_ia32_ktestzdi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestz_mask32_u8 (__mmask32 __A, __mmask32 __B)
{
  return (unsigned char) __builtin_ia32_ktestzsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestz_mask64_u8 (__mmask64 __A, __mmask64 __B)
{
  return (unsigned char) __builtin_ia32_ktestzdi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestc_mask32_u8 (__mmask32 __A, __mmask32 __B)
{
  return (unsigned char) __builtin_ia32_ktestcsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestc_mask64_u8 (__mmask64 __A, __mmask64 __B)
{
  return (unsigned char) __builtin_ia32_ktestcdi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortest_mask32_u8 (__mmask32 __A, __mmask32 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_kortestcsi (__A, __B);
  return (unsigned char) __builtin_ia32_kortestzsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortest_mask64_u8 (__mmask64 __A, __mmask64 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_kortestcdi (__A, __B);
  return (unsigned char) __builtin_ia32_kortestzdi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestz_mask32_u8 (__mmask32 __A, __mmask32 __B)
{
  return (unsigned char) __builtin_ia32_kortestzsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestz_mask64_u8 (__mmask64 __A, __mmask64 __B)
{
  return (unsigned char) __builtin_ia32_kortestzdi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestc_mask32_u8 (__mmask32 __A, __mmask32 __B)
{
  return (unsigned char) __builtin_ia32_kortestcsi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kortestc_mask64_u8 (__mmask64 __A, __mmask64 __B)
{
  return (unsigned char) __builtin_ia32_kortestcdi (__A, __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kadd_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kaddsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kadd_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kadddi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline unsigned int
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtmask32_u32 (__mmask32 __A)
{
  return (unsigned int) __builtin_ia32_kmovd ((__mmask32) __A);
}

extern __inline unsigned long long
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtmask64_u64 (__mmask64 __A)
{
  return (unsigned long long) __builtin_ia32_kmovq ((__mmask64) __A);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtu32_mask32 (unsigned int __A)
{
  return (__mmask32) __builtin_ia32_kmovd ((__mmask32) __A);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_cvtu64_mask64 (unsigned long long __A)
{
  return (__mmask64) __builtin_ia32_kmovq ((__mmask64) __A);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_load_mask32 (__mmask32 *__A)
{
  return (__mmask32) __builtin_ia32_kmovd (*__A);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_load_mask64 (__mmask64 *__A)
{
  return (__mmask64) __builtin_ia32_kmovq (*(__mmask64 *) __A);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_store_mask32 (__mmask32 *__A, __mmask32 __B)
{
  *(__mmask32 *) __A = __builtin_ia32_kmovd (__B);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_store_mask64 (__mmask64 *__A, __mmask64 __B)
{
  *(__mmask64 *) __A = __builtin_ia32_kmovq (__B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_knot_mask32 (__mmask32 __A)
{
  return (__mmask32) __builtin_ia32_knotsi ((__mmask32) __A);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_knot_mask64 (__mmask64 __A)
{
  return (__mmask64) __builtin_ia32_knotdi ((__mmask64) __A);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kor_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_korsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kor_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kordi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kxnor_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kxnorsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kxnor_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kxnordi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kxor_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kxorsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kxor_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kxordi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kand_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kandsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kand_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kanddi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kandn_mask32 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kandnsi ((__mmask32) __A, (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kandn_mask64 (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kandndi ((__mmask64) __A, (__mmask64) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_epi16 (__m512i __W, __mmask32 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdquhi512_mask ((__v32hi) __A,
          (__v32hi) __W,
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_epi16 (__mmask32 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdquhi512_mask ((__v32hi) __A,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_epi16 (__m512i __W, __mmask32 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddquhi512_mask ((const short *) __P,
           (__v32hi) __W,
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_epi16 (__mmask32 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddquhi512_mask ((const short *) __P,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_epi16 (void *__P, __mmask32 __U, __m512i __A)
{
  __builtin_ia32_storedquhi512_mask ((short *) __P,
         (__v32hi) __A,
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mov_epi8 (__m512i __W, __mmask64 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdquqi512_mask ((__v64qi) __A,
          (__v64qi) __W,
          (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mov_epi8 (__mmask64 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_movdquqi512_mask ((__v64qi) __A,
          (__v64qi)
          _mm512_setzero_si512 (),
          (__mmask64) __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kunpackw (__mmask32 __A, __mmask32 __B)
{
  return (__mmask32) __builtin_ia32_kunpcksi ((__mmask32) __A,
           (__mmask32) __B);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kunpackw_mask32 (__mmask16 __A, __mmask16 __B)
{
  return (__mmask32) __builtin_ia32_kunpcksi ((__mmask32) __A,
           (__mmask32) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_kunpackd (__mmask64 __A, __mmask64 __B)
{
  return (__mmask64) __builtin_ia32_kunpckdi ((__mmask64) __A,
           (__mmask64) __B);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_kunpackd_mask64 (__mmask32 __A, __mmask32 __B)
{
  return (__mmask64) __builtin_ia32_kunpckdi ((__mmask64) __A,
           (__mmask64) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_loadu_epi8 (__m512i __W, __mmask64 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddquqi512_mask ((const char *) __P,
           (__v64qi) __W,
           (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_loadu_epi8 (__mmask64 __U, void const *__P)
{
  return (__m512i) __builtin_ia32_loaddquqi512_mask ((const char *) __P,
           (__v64qi)
           _mm512_setzero_si512 (),
           (__mmask64) __U);
}

extern __inline void
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_storeu_epi8 (void *__P, __mmask64 __U, __m512i __A)
{
  __builtin_ia32_storedquqi512_mask ((char *) __P,
         (__v64qi) __A,
         (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sad_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psadbw512 ((__v64qi) __A,
          (__v64qi) __B);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi16_epi8 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
        (__v32qi) _mm256_undefined_si256(),
        (__mmask32) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
        (__v32qi) __O, __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi16_epi8 (__mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovwb512_mask ((__v32hi) __A,
        (__v32qi)
        _mm256_setzero_si256 (),
        __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtsepi16_epi8 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
         (__v32qi)_mm256_undefined_si256(),
         (__mmask32) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtsepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
         (__v32qi)__O,
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtsepi16_epi8 (__mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovswb512_mask ((__v32hi) __A,
         (__v32qi)
         _mm256_setzero_si256 (),
         __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtusepi16_epi8 (__m512i __A)
{
  return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
          (__v32qi)_mm256_undefined_si256(),
          (__mmask32) -1);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtusepi16_epi8 (__m256i __O, __mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
          (__v32qi) __O,
          __M);
}

extern __inline __m256i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtusepi16_epi8 (__mmask32 __M, __m512i __A)
{
  return (__m256i) __builtin_ia32_pmovuswb512_mask ((__v32hi) __A,
          (__v32qi)
          _mm256_setzero_si256 (),
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastb_epi8 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
             (__v64qi)_mm512_undefined_epi32(),
             (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastb_epi8 (__m512i __O, __mmask64 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
             (__v64qi) __O,
             __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastb_epi8 (__mmask64 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastb512_mask ((__v16qi) __A,
             (__v64qi)
             _mm512_setzero_si512 (),
             __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_set1_epi8 (__m512i __O, __mmask64 __M, char __A)
{
  return (__m512i) __builtin_ia32_pbroadcastb512_gpr_mask (__A,
          (__v64qi) __O,
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_set1_epi8 (__mmask64 __M, char __A)
{
  return (__m512i)
  __builtin_ia32_pbroadcastb512_gpr_mask (__A,
       (__v64qi)
       _mm512_setzero_si512 (),
       __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_broadcastw_epi16 (__m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
             (__v32hi)_mm512_undefined_epi32(),
             (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_broadcastw_epi16 (__m512i __O, __mmask32 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
             (__v32hi) __O,
             __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_broadcastw_epi16 (__mmask32 __M, __m128i __A)
{
  return (__m512i) __builtin_ia32_pbroadcastw512_mask ((__v8hi) __A,
             (__v32hi)
             _mm512_setzero_si512 (),
             __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_set1_epi16 (__m512i __O, __mmask32 __M, short __A)
{
  return (__m512i) __builtin_ia32_pbroadcastw512_gpr_mask (__A,
          (__v32hi) __O,
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_set1_epi16 (__mmask32 __M, short __A)
{
  return (__m512i)
  __builtin_ia32_pbroadcastw512_gpr_mask (__A,
       (__v32hi)
       _mm512_setzero_si512 (),
       __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mulhrs_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mulhrs_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
     __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v32hi) __W,
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mulhrs_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhrsw512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mulhi_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mulhi_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mulhi_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mulhi_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mulhi_epu16 (__m512i __W, __mmask32 __U, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi) __W,
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mulhi_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmulhuw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mullo_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v32hu) __A * (__v32hu) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_mullo_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_pmullw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_mullo_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmullw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepi8_epi16 (__m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbw512_mask ((__v32qi) __A,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepi8_epi16 (__m512i __W, __mmask32 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbw512_mask ((__v32qi) __A,
          (__v32hi) __W,
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepi8_epi16 (__mmask32 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovsxbw512_mask ((__v32qi) __A,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cvtepu8_epi16 (__m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbw512_mask ((__v32qi) __A,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cvtepu8_epi16 (__m512i __W, __mmask32 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbw512_mask ((__v32qi) __A,
          (__v32hi) __W,
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_cvtepu8_epi16 (__mmask32 __U, __m256i __A)
{
  return (__m512i) __builtin_ia32_pmovzxbw512_mask ((__v32qi) __A,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutexvar_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_permvarhi512_mask ((__v32hi) __B,
           (__v32hi) __A,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutexvar_epi16 (__mmask32 __M, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_permvarhi512_mask ((__v32hi) __B,
           (__v32hi) __A,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutexvar_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
          __m512i __B)
{
  return (__m512i) __builtin_ia32_permvarhi512_mask ((__v32hi) __B,
           (__v32hi) __A,
           (__v32hi) __W,
           (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_permutex2var_epi16 (__m512i __A, __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varhi512_mask ((__v32hi) __I
                 ,
       (__v32hi) __A,
       (__v32hi) __B,
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_permutex2var_epi16 (__m512i __A, __mmask32 __U,
    __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varhi512_mask ((__v32hi) __I
                 ,
       (__v32hi) __A,
       (__v32hi) __B,
       (__mmask32)
       __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask2_permutex2var_epi16 (__m512i __A, __m512i __I,
     __mmask32 __U, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermi2varhi512_mask ((__v32hi) __A,
       (__v32hi) __I
                 ,
       (__v32hi) __B,
       (__mmask32)
       __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_permutex2var_epi16 (__mmask32 __U, __m512i __A,
     __m512i __I, __m512i __B)
{
  return (__m512i) __builtin_ia32_vpermt2varhi512_maskz ((__v32hi) __I
                  ,
        (__v32hi) __A,
        (__v32hi) __B,
        (__mmask32)
        __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_avg_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_avg_epu8 (__m512i __W, __mmask64 __U, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi) __W,
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_avg_epu8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v64qu) __A + (__v64qu) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_paddb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi) __W,
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v64qu) __A - (__v64qu) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_psubb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi) __W,
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubb512_mask ((__v64qi) __A,
       (__v64qi) __B,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_avg_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_avg_epu16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_avg_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pavgw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_subs_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_subs_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_subs_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_subs_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi)
         _mm512_setzero_si512 (),
         (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_subs_epu8 (__m512i __W, __mmask64 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi) __W,
         (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_subs_epu8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi)
         _mm512_setzero_si512 (),
         (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_adds_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_adds_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_adds_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_adds_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi)
         _mm512_setzero_si512 (),
         (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_adds_epu8 (__m512i __W, __mmask64 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi) __W,
         (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_adds_epu8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusb512_mask ((__v64qi) __A,
         (__v64qi) __B,
         (__v64qi)
         _mm512_setzero_si512 (),
         (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sub_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v32hu) __A - (__v32hu) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sub_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_psubw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sub_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_subs_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_subs_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_subs_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_subs_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_subs_epu16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi) __W,
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_subs_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psubusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_add_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) ((__v32hu) __A + (__v32hu) __B);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_add_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_paddw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_add_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddw512_mask ((__v32hi) __A,
       (__v32hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_adds_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_adds_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_adds_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_adds_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_adds_epu16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi) __W,
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_adds_epu16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_paddusw512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v32hi)
         _mm512_setzero_si512 (),
         (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srl_epi16 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srl_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srl_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_packs_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packsswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi)
          _mm512_setzero_si512 (),
          (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sll_epi16 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sll_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m128i __B)
{
  return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sll_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psllw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maddubs_epi16 (__m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmaddubsw512_mask ((__v64qi) __X,
           (__v64qi) __Y,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_maddubs_epi16 (__m512i __W, __mmask32 __U, __m512i __X,
      __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmaddubsw512_mask ((__v64qi) __X,
           (__v64qi) __Y,
           (__v32hi) __W,
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_maddubs_epi16 (__mmask32 __U, __m512i __X, __m512i __Y)
{
  return (__m512i) __builtin_ia32_pmaddubsw512_mask ((__v64qi) __X,
           (__v64qi) __Y,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_madd_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaddwd512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v16si)
         _mm512_setzero_si512 (),
         (__mmask16) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_madd_epi16 (__m512i __W, __mmask16 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaddwd512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v16si) __W,
         (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_madd_epi16 (__mmask16 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaddwd512_mask ((__v32hi) __A,
         (__v32hi) __B,
         (__v16si)
         _mm512_setzero_si512 (),
         (__mmask16) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi)
           _mm512_setzero_si512 (),
           (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
      __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi) __W,
           (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi)
           _mm512_setzero_si512 (),
           (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpackhi_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpackhi_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
       __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi) __W,
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpackhi_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpckhwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi)
           _mm512_setzero_si512 (),
           (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
      __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi) __W,
           (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklbw512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__v64qi)
           _mm512_setzero_si512 (),
           (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_unpacklo_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_unpacklo_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
       __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi) __W,
           (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_unpacklo_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_punpcklwd512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__v32hi)
           _mm512_setzero_si512 (),
           (__mmask32) __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epu8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __A,
          (__v64qi) __B, 0,
          (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epi8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_pcmpeqb512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epu8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __A,
          (__v64qi) __B, 0,
          __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epi8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_pcmpeqb512_mask ((__v64qi) __A,
           (__v64qi) __B,
           __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epu16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __A,
          (__v32hi) __B, 0,
          (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpeq_epi16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_pcmpeqw512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epu16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __A,
          (__v32hi) __B, 0,
          __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpeq_epi16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_pcmpeqw512_mask ((__v32hi) __A,
           (__v32hi) __B,
           __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epu8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __A,
          (__v64qi) __B, 6,
          (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epi8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_pcmpgtb512_mask ((__v64qi) __A,
           (__v64qi) __B,
           (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epu8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __A,
          (__v64qi) __B, 6,
          __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epi8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_pcmpgtb512_mask ((__v64qi) __A,
           (__v64qi) __B,
           __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epu16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __A,
          (__v32hi) __B, 6,
          (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpgt_epi16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_pcmpgtw512_mask ((__v32hi) __A,
           (__v32hi) __B,
           (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epu16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __A,
          (__v32hi) __B, 6,
          __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpgt_epi16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_pcmpgtw512_mask ((__v32hi) __A,
           (__v32hi) __B,
           __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movepi8_mask (__m512i __A)
{
  return (__mmask64) __builtin_ia32_cvtb2mask512 ((__v64qi) __A);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movepi16_mask (__m512i __A)
{
  return (__mmask32) __builtin_ia32_cvtw2mask512 ((__v32hi) __A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movm_epi8 (__mmask64 __A)
{
  return (__m512i) __builtin_ia32_cvtmask2b512 (__A);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_movm_epi16 (__mmask32 __A)
{
  return (__m512i) __builtin_ia32_cvtmask2w512 (__A);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_test_epi8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ptestmb512 ((__v64qi) __A,
      (__v64qi) __B,
      (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_test_epi8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ptestmb512 ((__v64qi) __A,
      (__v64qi) __B, __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_test_epi16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ptestmw512 ((__v32hi) __A,
      (__v32hi) __B,
      (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_test_epi16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ptestmw512 ((__v32hi) __A,
      (__v32hi) __B, __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_testn_epi8_mask (__m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ptestnmb512 ((__v64qi) __A,
       (__v64qi) __B,
       (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_testn_epi8_mask (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__mmask64) __builtin_ia32_ptestnmb512 ((__v64qi) __A,
       (__v64qi) __B, __U);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_testn_epi16_mask (__m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ptestnmw512 ((__v32hi) __A,
       (__v32hi) __B,
       (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_testn_epi16_mask (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__mmask32) __builtin_ia32_ptestnmw512 ((__v32hi) __A,
       (__v32hi) __B, __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_shuffle_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pshufb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_shuffle_epi8 (__m512i __W, __mmask64 __U, __m512i __A,
     __m512i __B)
{
  return (__m512i) __builtin_ia32_pshufb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_shuffle_epi8 (__mmask64 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pshufb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epu16 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epu16 (__m512i __W, __mmask32 __M, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_pminuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epi16 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epu8 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epu8 (__m512i __W, __mmask64 __M, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epi8 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epi8 (__m512i __W, __mmask64 __M, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epu8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epu8 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epu8 (__m512i __W, __mmask64 __M, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_pminub512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_min_epi8 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_min_epi8 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi)
        _mm512_setzero_si512 (),
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_min_epi8 (__m512i __W, __mmask64 __M, __m512i __A,
        __m512i __B)
{
  return (__m512i) __builtin_ia32_pminsb512_mask ((__v64qi) __A,
        (__v64qi) __B,
        (__v64qi) __W,
        (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epi16 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxsw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_max_epu16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_max_epu16 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_max_epu16 (__m512i __W, __mmask32 __M, __m512i __A,
         __m512i __B)
{
  return (__m512i) __builtin_ia32_pmaxuw512_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sra_epi16 (__m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sra_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
         __m128i __B)
{
  return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sra_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
{
  return (__m512i) __builtin_ia32_psraw512_mask ((__v32hi) __A,
       (__v8hi) __B,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srav_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psrav32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srav_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_psrav32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srav_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psrav32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_srlv_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psrlv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_srlv_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_psrlv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_srlv_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psrlv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_sllv_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psllv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_sllv_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
   __m512i __B)
{
  return (__m512i) __builtin_ia32_psllv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi) __W,
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_sllv_epi16 (__mmask32 __U, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_psllv32hi_mask ((__v32hi) __A,
        (__v32hi) __B,
        (__v32hi)
        _mm512_setzero_si512 (),
        (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_packs_epi16 (__m512i __W, __mmask64 __M, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_packsswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi) __W,
          (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_packs_epi16 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packsswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi)
          _mm512_setzero_si512 (),
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_packus_epi16 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packuswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi)
          _mm512_setzero_si512 (),
          (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_packus_epi16 (__m512i __W, __mmask64 __M, __m512i __A,
     __m512i __B)
{
  return (__m512i) __builtin_ia32_packuswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi) __W,
          (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_packus_epi16 (__mmask64 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packuswb512_mask ((__v32hi) __A,
          (__v32hi) __B,
          (__v64qi)
          _mm512_setzero_si512 (),
          (__mmask64) __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_epi8 (__m512i __A)
{
  return (__m512i) __builtin_ia32_pabsb512_mask ((__v64qi) __A,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_epi8 (__m512i __W, __mmask64 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsb512_mask ((__v64qi) __A,
       (__v64qi) __W,
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_abs_epi8 (__mmask64 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsb512_mask ((__v64qi) __A,
       (__v64qi)
       _mm512_setzero_si512 (),
       (__mmask64) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_abs_epi16 (__m512i __A)
{
  return (__m512i) __builtin_ia32_pabsw512_mask ((__v32hi) __A,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_abs_epi16 (__m512i __W, __mmask32 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsw512_mask ((__v32hi) __A,
       (__v32hi) __W,
       (__mmask32) __U);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_abs_epi16 (__mmask32 __U, __m512i __A)
{
  return (__m512i) __builtin_ia32_pabsw512_mask ((__v32hi) __A,
       (__v32hi)
       _mm512_setzero_si512 (),
       (__mmask32) __U);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epu8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 4,
         (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epu8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 1,
         (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epu8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 5,
         (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epu8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 2,
         (__mmask64) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epu16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 4,
         (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epu16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 1,
         (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epu16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 5,
         (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epu16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 2,
         (__mmask32) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epi8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 4,
        (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epi8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 1,
        (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epi8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 5,
        (__mmask64) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epi8_mask (__mmask64 __M, __m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 2,
        (__mmask64) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpneq_epi16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 4,
        (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmplt_epi16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 1,
        (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmpge_epi16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 5,
        (__mmask32) __M);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_cmple_epi16_mask (__mmask32 __M, __m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 2,
        (__mmask32) __M);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 4,
         (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 1,
         (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 5,
         (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_ucmpb512_mask ((__v64qi) __X,
         (__v64qi) __Y, 2,
         (__mmask64) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epu16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 4,
         (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epu16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 1,
         (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epu16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 5,
         (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epu16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_ucmpw512_mask ((__v32hi) __X,
         (__v32hi) __Y, 2,
         (__mmask32) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epi8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 4,
        (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 1,
        (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epi8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 5,
        (__mmask64) -1);
}

extern __inline __mmask64
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi8_mask (__m512i __X, __m512i __Y)
{
  return (__mmask64) __builtin_ia32_cmpb512_mask ((__v64qi) __X,
        (__v64qi) __Y, 2,
        (__mmask64) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpneq_epi16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 4,
        (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmplt_epi16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 1,
        (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmpge_epi16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 5,
        (__mmask32) -1);
}

extern __inline __mmask32
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_cmple_epi16_mask (__m512i __X, __m512i __Y)
{
  return (__mmask32) __builtin_ia32_cmpw512_mask ((__v32hi) __X,
        (__v32hi) __Y, 2,
        (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_packs_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packssdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_packs_epi32 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packssdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_packs_epi32 (__m512i __W, __mmask32 __M, __m512i __A,
    __m512i __B)
{
  return (__m512i) __builtin_ia32_packssdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi) __W,
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_packus_epi32 (__m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packusdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          (__mmask32) -1);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_maskz_packus_epi32 (__mmask32 __M, __m512i __A, __m512i __B)
{
  return (__m512i) __builtin_ia32_packusdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi)
          _mm512_setzero_si512 (),
          __M);
}

extern __inline __m512i
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_mm512_mask_packus_epi32 (__m512i __W, __mmask32 __M, __m512i __A,
     __m512i __B)
{
  return (__m512i) __builtin_ia32_packusdw512_mask ((__v16si) __A,
          (__v16si) __B,
          (__v32hi) __W,
          __M);
}
# 3277 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512bwintrin.h" 3 4
#pragma GCC pop_options
# 56 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/immintrin.h" 2 3 4

# 1 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512dqintrin.h" 1 3 4
# 32 "C:/msys64/mingw64/lib/gcc/x86_64-w64-mingw32/7.3.0/include/avx512dqintrin.h" 3 4
#pragma GCC push_options
#pragma GCC target("avx512dq")



extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktest_mask8_u8 (__mmask8 __A, __mmask8 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_ktestcqi (__A, __B);
  return (unsigned char) __builtin_ia32_ktestzqi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestz_mask8_u8 (__mmask8 __A, __mmask8 __B)
{
  return (unsigned char) __builtin_ia32_ktestzqi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestc_mask8_u8 (__mmask8 __A, __mmask8 __B)
{
  return (unsigned char) __builtin_ia32_ktestcqi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktest_mask16_u8 (__mmask16 __A, __mmask16 __B, unsigned char *__CF)
{
  *__CF = (unsigned char) __builtin_ia32_ktestchi (__A, __B);
  return (unsigned char) __builtin_ia32_ktestzhi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
_ktestz_mask16_u8 (__mmask16 __A, __mmask16 __B)
{
  return (unsigned char) __builtin_ia32_ktestzhi (__A, __B);
}

extern __inline unsigned char
__attribute__ ((__gnu_inline__,