ivec.h

C语言库函数的原型,有用的拿去

/***
*** Copyright (C) 1985-1999 Intel Corporation.  All rights reserved.
***
*** The information and source code contained herein is the exclusive
*** property of Intel Corporation and may not be disclosed, examined
*** or reproduced in whole or in part without explicit written authorization
*** from the company.
***
****/

/*
 *  Definition of a C++ class interface to MMX(TM) instruction intrinsics.
 *
 */

#ifndef _IVEC_H_INCLUDED
#define _IVEC_H_INCLUDED
#ifndef RC_INVOKED

#if !defined __cplusplus
        #error ERROR: This file is only supported in C++ compilations!
#endif  /* !defined __cplusplus */

#if defined (_M_CEE_PURE)
        #error ERROR: This file is not supported in the pure mode!
#else  /* defined (_M_CEE_PURE) */

#include <mmintrin.h>

#ifndef _VEC_ASSERT
    #include <crtdefs.h>

#ifdef NDEBUG
        #define _VEC_ASSERT(_Expression) ((void)0)
#else  /* NDEBUG */
#ifdef __cplusplus
            extern "C" {
#endif  /* __cplusplus */

        _CRTIMP void __cdecl _wassert(_In_z_ const wchar_t * _Message, _In_z_ const wchar_t *_File, _In_ unsigned _Line);

#ifdef __cplusplus
            }
#endif  /* __cplusplus */

        #define _VEC_ASSERT(_Expression) (void)( (!!(_Expression)) || (_wassert(_CRT_WIDE(#_Expression), _CRT_WIDE(__FILE__), __LINE__), 0) )
#endif  /* NDEBUG */
#endif  /* _VEC_ASSERT */

/*
 * Define _SILENCE_IVEC_C4799 to disable warning C4799 inside this header.
 * Be careful that any code that uses these functions properly executes EMMS
 * or _m_empty() after using any MMX instruction and before using the x87 NDP.
 */
#if defined (_SILENCE_IVEC_C4799)
        #pragma warning(push)
        #pragma warning(disable: 4799)
#endif  /* defined (_SILENCE_IVEC_C4799) */

/*
 * Define _ENABLE_VEC_DEBUG to enable std::ostream inserters for debug output
 */
#if defined (_ENABLE_VEC_DEBUG)
        #include <iostream>
#endif  /* defined (_ENABLE_VEC_DEBUG) */

class I8vec8;                   /* 8 elements, each element a signed or unsigned char data type */
class Is8vec8;                  /* 8 elements, each element a signed char data type */
class Iu8vec8;                  /* 8 elements, each element an unsigned char data type */
class I16vec4;                  /* 4 elements, each element a signed or unsigned short */
class Is16vec4;                 /* 4 elements, each element a signed short */
class Iu16vec4;                 /* 4 elements, each element an unsigned short */
class I32vec2;                  /* 2 elements, each element a signed or unsigned long */
class Is32vec2;                 /* 2 elements, each element a signed long */
class Iu32vec2;                 /* 2 elements, each element a unsigned long */
class I64vec1;                  /* 1 element, a __m64 data type - Base I64vec1 class  */

#define _MM_8UB(element,vector) (*((unsigned char*)&##vector + ##element))
#define _MM_8B(element,vector) (*((signed char*)&##vector + ##element))

#define _MM_4UW(element,vector) (*((unsigned short*)&##vector + ##element))
#define _MM_4W(element,vector) (*((short*)&##vector + ##element))

#define _MM_2UDW(element,vector) (*((unsigned int*)&##vector + ##element))
#define _MM_2DW(element,vector) (*((int*)&##vector + ##element))

#define _MM_QW (*((__int64*)&vec))

/* M64 Class:
 * 1 element, a __m64 data type
 * Contructors & Logical Operations
 */
class M64
{
protected:
                __m64 vec;

public:
        M64()                                                                   { }
        M64(__m64 mm)                                                   { vec = mm; }
        M64(__int64 mm)                                                 { _MM_QW = mm; }
        M64(int i)                                                              { vec = _m_from_int(i); }

        operator __m64() const                                  { return vec; }

        /* Logical Operations */
        M64& operator&=(const M64 &a)                                   { return *this = (M64) _m_pand(vec,a); }
        M64& operator|=(const M64 &a)                                   { return *this = (M64) _m_por(vec,a); }
        M64& operator^=(const M64 &a)                                   { return *this = (M64) _m_pxor(vec,a); }

};

inline M64 operator&(const M64 &a, const M64 &b)        { return _m_pand( a,b); }
inline M64 operator|(const M64 &a, const M64 &b)        { return _m_por(a,b); }
inline M64 operator^(const M64 &a, const M64 &b)        { return _m_pxor(a,b); }
inline M64 andnot(const M64 &a, const M64 &b)           { return _m_pandn(a,b); }

/* I64vec1 Class:
 * 1 element, a __m64 data type
 * Contains Operations which can operate on any __m64 data type
 */

class I64vec1 : public M64
{
public:
        I64vec1()                                                               { }
        I64vec1(__m64 mm) : M64(mm)                             { }
        explicit I64vec1(int i) : M64(i)                { }
        explicit I64vec1(__int64 mm) : M64(mm)  { }

        I64vec1& operator= (const M64 &a) { return *this = (I64vec1) a; }
        I64vec1& operator&=(const M64 &a) { return *this = (I64vec1) _m_pand(vec,a); }
        I64vec1& operator|=(const M64 &a) { return *this = (I64vec1) _m_por(vec,a); }
        I64vec1& operator^=(const M64 &a) { return *this = (I64vec1) _m_pxor(vec,a); }

        /* Shift Logical Operations */
        I64vec1 operator<<(const M64 &a)                                { return _m_psllq(vec, a); }
        I64vec1 operator<<(int count)                               { return _m_psllqi(vec, count); }
        I64vec1& operator<<=(const M64 &a)                              { return *this = (I64vec1) _m_psllq(vec, a); }
        I64vec1& operator<<=(int count)                                 { return *this = (I64vec1) _m_psllqi(vec, count); }
        I64vec1 operator>>(const M64 &a)                                { return _m_psrlq(vec, a); }
        I64vec1 operator>>(int count)                                   { return _m_psrlqi(vec, count); }
        I64vec1& operator>>=(const M64 &a)                              { return *this = (I64vec1) _m_psrlq(vec, a); }
        I64vec1& operator>>=(int count)                                 { return *this = (I64vec1) _m_psrlqi(vec, count); }
};

/* I32vec2 Class:
 * 2 elements, each element either a signed or unsigned int
 */
class I32vec2 : public M64
{
public:
        I32vec2() { }
        I32vec2(__m64 mm) : M64(mm) { }
        explicit I32vec2(int i) : M64 (i) { }
        explicit I32vec2(__int64 i): M64(i) {}

        /* Assignment Operator */
        I32vec2& operator= (const M64 &a) { return *this = (I32vec2) a; }

        /* Logical Assignment Operators */
        I32vec2& operator&=(const M64 &a) { return *this = (I32vec2) _m_pand(vec,a); }
        I32vec2& operator|=(const M64 &a) { return *this = (I32vec2) _m_por(vec,a); }
        I32vec2& operator^=(const M64 &a) { return *this = (I32vec2) _m_pxor(vec,a); }

        /* Addition & Subtraction Assignment Operators */
        I32vec2& operator +=(const I32vec2 &a)                  { return *this = (I32vec2) _m_paddd(vec,a); }
        I32vec2& operator -=(const I32vec2 &a)                  { return *this = (I32vec2) _m_psubd(vec,a); }

        /* Shift Logical Operators */
        I32vec2 operator<<(const I32vec2 &a)                    { return _m_pslld(vec,a); }
        I32vec2 operator<<(int count)                               { return _m_pslldi(vec,count); }
        I32vec2& operator<<=(const I32vec2 &a)                  { return *this = (I32vec2) _m_pslld(vec,a); }
        I32vec2& operator<<=(int count)                                 { return *this = (I32vec2) _m_pslldi(vec,count); }

};

/* Compare For Equality */
inline I32vec2 cmpeq(const I32vec2 &a, const I32vec2 &b)            { return _m_pcmpeqd(a,b); }
inline I32vec2 cmpneq(const I32vec2 &a, const I32vec2 &b)           { return _m_pandn(_m_pcmpeqd(a,b), M64(0xffffffffffffffffi64)); }
/* Unpacks */
inline I32vec2 unpack_low(const I32vec2 &a, const I32vec2 &b)   {return _m_punpckldq(a,b); }
inline I32vec2 unpack_high(const I32vec2 &a, const I32vec2 &b)  {return _m_punpckhdq(a,b); }

/* Is32vec2 Class:
 * 2 elements, each element a signed int
 */
class Is32vec2 : public I32vec2
{
public:
        Is32vec2() { }
        Is32vec2(__m64 mm) : I32vec2(mm) { }
        Is32vec2(signed int i0, signed int i1)
        {
                _MM_2DW(0,vec) = i1;
                _MM_2DW(1,vec) = i0;
        }
        explicit Is32vec2(int i) : I32vec2 (i)          {}
        explicit Is32vec2(__int64 i): I32vec2(i)        {}

        /* Assignment Operator */
        Is32vec2& operator= (const M64 &a)              { return *this = (Is32vec2) a; }

        /* Logical Assignment Operators */
        Is32vec2& operator&=(const M64 &a)              { return *this = (Is32vec2) _m_pand(vec,a); }
        Is32vec2& operator|=(const M64 &a)              { return *this = (Is32vec2) _m_por(vec,a); }
        Is32vec2& operator^=(const M64 &a)              { return *this = (Is32vec2) _m_pxor(vec,a); }

        /* Addition & Subtraction Assignment Operators */
        Is32vec2& operator +=(const I32vec2 &a) { return *this = (Is32vec2) _m_paddd(vec,a); }
        Is32vec2& operator -=(const I32vec2 &a) { return *this = (Is32vec2) _m_psubd(vec,a); }

        /* Shift Logical Operators */
        Is32vec2 operator<<(const M64 &a)               { return _m_pslld(vec,a); }
        Is32vec2 operator<<(int count)                  { return _m_pslldi(vec,count); }
        Is32vec2& operator<<=(const M64 &a)             { return *this = (Is32vec2) _m_pslld(vec,a); }
        Is32vec2& operator<<=(int count)                { return *this = (Is32vec2) _m_pslldi(vec,count); }
        /* Shift Arithmetic Operations */
        Is32vec2 operator>>(const M64 &a)               { return _m_psrad(vec, a); }
        Is32vec2 operator>>(int count)                  { return _m_psradi(vec, count); }
        Is32vec2& operator>>=(const M64 &a)             { return *this = (Is32vec2) _m_psrad(vec, a); }
        Is32vec2& operator>>=(int count)                { return *this = (Is32vec2) _m_psradi(vec, count); }

#if defined (_ENABLE_VEC_DEBUG)
        /* Output for Debug */
        friend std::ostream& operator<< (std::ostream &os, const Is32vec2 &a)
        {
                os << " [1]:" << _MM_2DW(1,a)
                << " [0]:" << _MM_2DW(0,a);
                return os;
        }
#endif  /* defined (_ENABLE_VEC_DEBUG) */

        /* Element Access for Debug, No data modified */
        const int& operator[](int i)const
        {
                _VEC_ASSERT(static_cast<unsigned int>(i) < 2);  /* Only 2 elements to access */
                return _MM_2DW(i,vec);
        }

        /* Element Access and Assignment for Debug */
        int& operator[](int i)
        {
                _VEC_ASSERT(static_cast<unsigned int>(i) < 2);  /* Only 2 elements to access */
                return _MM_2DW(i,vec);
        }
};

/* Compares */
inline Is32vec2 cmpeq(const Is32vec2 &a, const Is32vec2 &b)         { return _m_pcmpeqd(a,b); }
inline Is32vec2 cmpneq(const Is32vec2 &a, const Is32vec2 &b)        { return _m_pandn(_m_pcmpeqd(a,b), M64(0xffffffffffffffffi64)); }
inline Is32vec2 cmpgt(const Is32vec2 &a, const Is32vec2 &b)                     { return _m_pcmpgtd(a,b); }
inline Is32vec2 cmplt(const Is32vec2 &a, const Is32vec2 &b)                     { return _m_pcmpgtd(b,a); }
inline Is32vec2 cmple(const Is32vec2 &a, const Is32vec2 &b)                     { return _m_pandn(_m_pcmpgtd(a,b), M64(0xffffffffffffffffi64)); }
inline Is32vec2 cmpge(const Is32vec2 &a, const Is32vec2 &b)                     { return _m_pandn(_m_pcmpgtd(b,a), M64(0xffffffffffffffffi64)); }
/* Unpacks & Pack */
inline Is32vec2 unpack_low(const Is32vec2 &a, const Is32vec2 &b)        { return _m_punpckldq(a,b); }
inline Is32vec2 unpack_high(const Is32vec2 &a, const Is32vec2 &b)       { return _m_punpckhdq(a,b); }

/* Iu32vec2 Class:
 * 2 elements, each element unsigned int
 */
class Iu32vec2 : public I32vec2
{
public:
        Iu32vec2() { }
        Iu32vec2(__m64 mm) : I32vec2(mm) { }
        Iu32vec2(unsigned int ui0, unsigned int ui1)
        {
                _MM_2UDW(0,vec) = ui1;
                _MM_2UDW(1,vec) = ui0;
        }

        explicit Iu32vec2(int i) : I32vec2 (i)          { }
        explicit Iu32vec2(__int64 i) : I32vec2 (i)      { }

        /* Assignment Operator */
        Iu32vec2& operator= (const M64 &a)              { return *this = (Iu32vec2) a; }

        /* Logical Assignment Operators */
        Iu32vec2& operator&=(const M64 &a)              { return *this = (Iu32vec2) _m_pand(vec,a); }
        Iu32vec2& operator|=(const M64 &a)              { return *this = (Iu32vec2) _m_por(vec,a); }
        Iu32vec2& operator^=(const M64 &a)              { return *this = (Iu32vec2) _m_pxor(vec,a); }

        /* Addition & Subtraction Assignment Operators */