1631 lines
68 KiB
C
1631 lines
68 KiB
C
|
/***
|
||
|
*** Copyright (C) 1985-2011 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 Intel(R) Pentium(R) 4 processor SSE2 intrinsics.
|
||
|
*
|
||
|
* File name : dvec.h class definitions
|
||
|
*
|
||
|
* Concept: A C++ abstraction of Intel(R) Pentium(R) 4 processor SSE2
|
||
|
* designed to improve programmer productivity. Speed and accuracy are
|
||
|
* sacrificed for utility. Facilitates an easy transition to compiler
|
||
|
* intrinsics or assembly language.
|
||
|
*
|
||
|
*/
|
||
|
|
||
|
#ifndef _DVEC_H_INCLUDED
|
||
|
#define _DVEC_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 <immintrin.h> /* SSE2 intrinsic function definition include file */
|
||
|
#include <fvec.h>
|
||
|
#include <crtdefs.h>
|
||
|
|
||
|
#ifndef _VEC_ASSERT
|
||
|
#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 */
|
||
|
|
||
|
#ifdef _MSC_VER
|
||
|
#pragma pack(push,_CRT_PACKING)
|
||
|
#endif /* _MSC_VER */
|
||
|
|
||
|
/* Define _ENABLE_VEC_DEBUG to enable std::ostream inserters for debug output */
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
#include <iostream>
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
#pragma pack(push,16) /* Must ensure class & union 16-B aligned */
|
||
|
|
||
|
const union
|
||
|
{
|
||
|
int i[4];
|
||
|
__m128d m;
|
||
|
} __f64vec2_abs_mask_cheat = {0xffffffff, 0x7fffffff, 0xffffffff, 0x7fffffff};
|
||
|
|
||
|
#define _f64vec2_abs_mask ((F64vec2)__f64vec2_abs_mask_cheat.m)
|
||
|
|
||
|
/* EMM Functionality Intrinsics */
|
||
|
|
||
|
class I8vec16; /* 16 elements, each element a signed or unsigned char data type */
|
||
|
class Is8vec16; /* 16 elements, each element a signed char data type */
|
||
|
class Iu8vec16; /* 16 elements, each element an unsigned char data type */
|
||
|
class I16vec8; /* 8 elements, each element a signed or unsigned short */
|
||
|
class Is16vec8; /* 8 elements, each element a signed short */
|
||
|
class Iu16vec8; /* 8 elements, each element an unsigned short */
|
||
|
class I32vec4; /* 4 elements, each element a signed or unsigned long */
|
||
|
class Is32vec4; /* 4 elements, each element a signed long */
|
||
|
class Iu32vec4; /* 4 elements, each element a unsigned long */
|
||
|
class I64vec2; /* 2 element, each a __m64 data type */
|
||
|
class I128vec1; /* 1 element, a __m128i data type */
|
||
|
|
||
|
#define _MM_16UB(element,vector) (*((unsigned char*)&##vector + ##element))
|
||
|
#define _MM_16B(element,vector) (*((signed char*)&##vector + ##element))
|
||
|
|
||
|
#define _MM_8UW(element,vector) (*((unsigned short*)&##vector + ##element))
|
||
|
#define _MM_8W(element,vector) (*((short*)&##vector + ##element))
|
||
|
|
||
|
#define _MM_4UDW(element,vector) (*((unsigned int*)&##vector + ##element))
|
||
|
#define _MM_4DW(element,vector) (*((int*)&##vector + ##element))
|
||
|
|
||
|
#define _MM_2QW(element,vector) (*((__int64*)&##vector + ##element))
|
||
|
|
||
|
|
||
|
/* We need a m128i constant, keeping performance in mind*/
|
||
|
|
||
|
#pragma warning(push)
|
||
|
#pragma warning(disable : 4640)
|
||
|
inline const __m128i get_mask128()
|
||
|
{
|
||
|
static const __m128i mask128 = _mm_set1_epi64(M64(0xffffffffffffffffi64));
|
||
|
return mask128;
|
||
|
}
|
||
|
#pragma warning(pop)
|
||
|
|
||
|
|
||
|
//DEVDIV Remove alais created in public\sdk\inc\winnt.h
|
||
|
#ifdef M128
|
||
|
#undef M128
|
||
|
#endif /* M128 */
|
||
|
#ifdef PM128
|
||
|
#undef PM128
|
||
|
#endif /* PM128 */
|
||
|
//end DEVDIV
|
||
|
|
||
|
/* M128 Class:
|
||
|
* 1 element, a __m128i data type
|
||
|
* Contructors & Logical Operations
|
||
|
*/
|
||
|
|
||
|
class M128
|
||
|
{
|
||
|
protected:
|
||
|
__m128i vec;
|
||
|
|
||
|
public:
|
||
|
M128() { }
|
||
|
M128(__m128i mm) { vec = mm; }
|
||
|
|
||
|
operator __m128i() const { return vec; }
|
||
|
|
||
|
/* Logical Operations */
|
||
|
M128& operator&=(const M128 &a) { return *this = (M128) _mm_and_si128(vec,a); }
|
||
|
M128& operator|=(const M128 &a) { return *this = (M128) _mm_or_si128(vec,a); }
|
||
|
M128& operator^=(const M128 &a) { return *this = (M128) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
};
|
||
|
|
||
|
inline M128 operator&(const M128 &a, const M128 &b) { return _mm_and_si128(a,b); }
|
||
|
inline M128 operator|(const M128 &a, const M128 &b) { return _mm_or_si128(a,b); }
|
||
|
inline M128 operator^(const M128 &a, const M128 &b) { return _mm_xor_si128(a,b); }
|
||
|
inline M128 andnot(const M128 &a, const M128 &b) { return _mm_andnot_si128(a,b); }
|
||
|
|
||
|
/* I128vec1 Class:
|
||
|
* 1 element, a __m128i data type
|
||
|
* Contains Operations which can operate on any __m6128i data type
|
||
|
*/
|
||
|
|
||
|
class I128vec1 : public M128
|
||
|
{
|
||
|
public:
|
||
|
I128vec1() { }
|
||
|
I128vec1(__m128i mm) : M128(mm) { }
|
||
|
|
||
|
I128vec1& operator= (const M128 &a) { return *this = (I128vec1) a; }
|
||
|
I128vec1& operator&=(const M128 &a) { return *this = (I128vec1) _mm_and_si128(vec,a); }
|
||
|
I128vec1& operator|=(const M128 &a) { return *this = (I128vec1) _mm_or_si128(vec,a); }
|
||
|
I128vec1& operator^=(const M128 &a) { return *this = (I128vec1) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
};
|
||
|
|
||
|
/* I64vec2 Class:
|
||
|
* 2 elements, each element signed or unsigned 64-bit integer
|
||
|
*/
|
||
|
class I64vec2 : public M128
|
||
|
{
|
||
|
public:
|
||
|
I64vec2() { }
|
||
|
I64vec2(__m128i mm) : M128(mm) { }
|
||
|
|
||
|
I64vec2(__m64 q1, __m64 q0)
|
||
|
{
|
||
|
_MM_2QW(0,vec) = *(__int64*)&q0;
|
||
|
_MM_2QW(1,vec) = *(__int64*)&q1;
|
||
|
}
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
I64vec2& operator= (const M128 &a) { return *this = (I64vec2) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
I64vec2& operator&=(const M128 &a) { return *this = (I64vec2) _mm_and_si128(vec,a); }
|
||
|
I64vec2& operator|=(const M128 &a) { return *this = (I64vec2) _mm_or_si128(vec,a); }
|
||
|
I64vec2& operator^=(const M128 &a) { return *this = (I64vec2) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
I64vec2& operator +=(const I64vec2 &a) { return *this = (I64vec2) _mm_add_epi64(vec,a); }
|
||
|
I64vec2& operator -=(const I64vec2 &a) { return *this = (I64vec2) _mm_sub_epi64(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
I64vec2 operator<<(const I64vec2 &a) { return _mm_sll_epi64(vec,a); }
|
||
|
I64vec2 operator<<(int count) { return _mm_slli_epi64(vec,count); }
|
||
|
I64vec2& operator<<=(const I64vec2 &a) { return *this = (I64vec2) _mm_sll_epi64(vec,a); }
|
||
|
I64vec2& operator<<=(int count) { return *this = (I64vec2) _mm_slli_epi64(vec,count); }
|
||
|
I64vec2 operator>>(const I64vec2 &a) { return _mm_srl_epi64(vec,a); }
|
||
|
I64vec2 operator>>(int count) { return _mm_srli_epi64(vec,count); }
|
||
|
I64vec2& operator>>=(const I64vec2 &a) { return *this = (I64vec2) _mm_srl_epi64(vec,a); }
|
||
|
I64vec2& operator>>=(int count) { return *this = (I64vec2) _mm_srli_epi64(vec,count); }
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const __int64& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 2); /* Only 2 elements to access */
|
||
|
return _MM_2QW(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access and Assignment for Debug */
|
||
|
__int64& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 2); /* Only 2 elements to access */
|
||
|
return _MM_2QW(i,vec);
|
||
|
}
|
||
|
|
||
|
|
||
|
};
|
||
|
|
||
|
/* Unpacks */
|
||
|
inline I64vec2 unpack_low(const I64vec2 &a, const I64vec2 &b) {return _mm_unpacklo_epi64(a,b); }
|
||
|
inline I64vec2 unpack_high(const I64vec2 &a, const I64vec2 &b) {return _mm_unpackhi_epi64(a,b); }
|
||
|
|
||
|
/* I32vec4 Class:
|
||
|
* 4 elements, each element either a signed or unsigned int
|
||
|
*/
|
||
|
class I32vec4 : public M128
|
||
|
{
|
||
|
public:
|
||
|
I32vec4() { }
|
||
|
I32vec4(__m128i mm) : M128(mm) { }
|
||
|
I32vec4(int i3, int i2, int i1, int i0) {vec = _mm_set_epi32(i3, i2, i1, i0);}
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
I32vec4& operator= (const M128 &a) { return *this = (I32vec4) a; }
|
||
|
|
||
|
/* Logicals Operators */
|
||
|
I32vec4& operator&=(const M128 &a) { return *this = (I32vec4) _mm_and_si128(vec,a); }
|
||
|
I32vec4& operator|=(const M128 &a) { return *this = (I32vec4) _mm_or_si128(vec,a); }
|
||
|
I32vec4& operator^=(const M128 &a) { return *this = (I32vec4) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
I32vec4& operator +=(const I32vec4 &a) { return *this = (I32vec4)_mm_add_epi32(vec,a); }
|
||
|
I32vec4& operator -=(const I32vec4 &a) { return *this = (I32vec4)_mm_sub_epi32(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
I32vec4 operator<<(const I32vec4 &a) { return _mm_sll_epi32(vec,a); }
|
||
|
I32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
|
||
|
I32vec4& operator<<=(const I32vec4 &a) { return *this = (I32vec4)_mm_sll_epi32(vec,a); }
|
||
|
I32vec4& operator<<=(int count) { return *this = (I32vec4)_mm_slli_epi32(vec,count); }
|
||
|
|
||
|
};
|
||
|
|
||
|
inline I32vec4 cmpeq(const I32vec4 &a, const I32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
|
||
|
inline I32vec4 cmpneq(const I32vec4 &a, const I32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b), get_mask128()); }
|
||
|
|
||
|
inline I32vec4 unpack_low(const I32vec4 &a, const I32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
|
||
|
inline I32vec4 unpack_high(const I32vec4 &a, const I32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
|
||
|
|
||
|
/* Is32vec4 Class:
|
||
|
* 4 elements, each element signed integer
|
||
|
*/
|
||
|
class Is32vec4 : public I32vec4
|
||
|
{
|
||
|
public:
|
||
|
Is32vec4() { }
|
||
|
Is32vec4(__m128i mm) : I32vec4(mm) { }
|
||
|
Is32vec4(int i3, int i2, int i1, int i0) : I32vec4(i3, i2, i1, i0){}
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Is32vec4& operator= (const M128 &a) { return *this = (Is32vec4) a; }
|
||
|
|
||
|
/* Logical Operators */
|
||
|
Is32vec4& operator&=(const M128 &a) { return *this = (Is32vec4) _mm_and_si128(vec,a); }
|
||
|
Is32vec4& operator|=(const M128 &a) { return *this = (Is32vec4) _mm_or_si128(vec,a); }
|
||
|
Is32vec4& operator^=(const M128 &a) { return *this = (Is32vec4) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Is32vec4& operator +=(const I32vec4 &a) { return *this = (Is32vec4)_mm_add_epi32(vec,a); }
|
||
|
Is32vec4& operator -=(const I32vec4 &a) { return *this = (Is32vec4)_mm_sub_epi32(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
Is32vec4 operator<<(const M128 &a) { return _mm_sll_epi32(vec,a); }
|
||
|
Is32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
|
||
|
Is32vec4& operator<<=(const M128 &a) { return *this = (Is32vec4)_mm_sll_epi32(vec,a); }
|
||
|
Is32vec4& operator<<=(int count) { return *this = (Is32vec4)_mm_slli_epi32(vec,count); }
|
||
|
/* Shift Arithmetic Operations */
|
||
|
Is32vec4 operator>>(const M128 &a) { return _mm_sra_epi32(vec,a); }
|
||
|
Is32vec4 operator>>(int count) { return _mm_srai_epi32(vec,count); }
|
||
|
Is32vec4& operator>>=(const M128 &a) { return *this = (Is32vec4) _mm_sra_epi32(vec,a); }
|
||
|
Is32vec4& operator>>=(int count) { return *this = (Is32vec4) _mm_srai_epi32(vec,count); }
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator<< (std::ostream &os, const Is32vec4 &a)
|
||
|
{
|
||
|
os << "[3]:" << _MM_4DW(3,a)
|
||
|
<< " [2]:" << _MM_4DW(2,a)
|
||
|
<< " [1]:" << _MM_4DW(1,a)
|
||
|
<< " [0]:" << _MM_4DW(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) < 4); /* Only 4 elements to access */
|
||
|
return _MM_4DW(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access for Debug */
|
||
|
int& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 4); /* Only 4 elements to access */
|
||
|
return _MM_4DW(i,vec);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
/* Compares */
|
||
|
inline Is32vec4 cmpeq(const Is32vec4 &a, const Is32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
|
||
|
inline Is32vec4 cmpneq(const Is32vec4 &a, const Is32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b), get_mask128()); }
|
||
|
inline Is32vec4 cmpgt(const Is32vec4 &a, const Is32vec4 &b) { return _mm_cmpgt_epi32(a,b); }
|
||
|
inline Is32vec4 cmplt(const Is32vec4 &a, const Is32vec4 &b) { return _mm_cmpgt_epi32(b,a); }
|
||
|
|
||
|
/* Unpacks */
|
||
|
inline Is32vec4 unpack_low(const Is32vec4 &a, const Is32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
|
||
|
inline Is32vec4 unpack_high(const Is32vec4 &a, const Is32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
|
||
|
|
||
|
/* Iu32vec4 Class:
|
||
|
* 4 elements, each element unsigned int
|
||
|
*/
|
||
|
class Iu32vec4 : public I32vec4
|
||
|
{
|
||
|
public:
|
||
|
Iu32vec4() { }
|
||
|
Iu32vec4(__m128i mm) : I32vec4(mm) { }
|
||
|
Iu32vec4(unsigned int ui3, unsigned int ui2, unsigned int ui1, unsigned int ui0)
|
||
|
: I32vec4(ui3, ui2, ui1, ui0) { }
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Iu32vec4& operator= (const M128 &a) { return *this = (Iu32vec4) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
Iu32vec4& operator&=(const M128 &a) { return *this = (Iu32vec4) _mm_and_si128(vec,a); }
|
||
|
Iu32vec4& operator|=(const M128 &a) { return *this = (Iu32vec4) _mm_or_si128(vec,a); }
|
||
|
Iu32vec4& operator^=(const M128 &a) { return *this = (Iu32vec4) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Iu32vec4& operator +=(const I32vec4 &a) { return *this = (Iu32vec4)_mm_add_epi32(vec,a); }
|
||
|
Iu32vec4& operator -=(const I32vec4 &a) { return *this = (Iu32vec4)_mm_sub_epi32(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
Iu32vec4 operator<<(const M128 &a) { return _mm_sll_epi32(vec,a); }
|
||
|
Iu32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
|
||
|
Iu32vec4& operator<<=(const M128 &a) { return *this = (Iu32vec4)_mm_sll_epi32(vec,a); }
|
||
|
Iu32vec4& operator<<=(int count) { return *this = (Iu32vec4)_mm_slli_epi32(vec,count); }
|
||
|
Iu32vec4 operator>>(const M128 &a) { return _mm_srl_epi32(vec,a); }
|
||
|
Iu32vec4 operator>>(int count) { return _mm_srli_epi32(vec,count); }
|
||
|
Iu32vec4& operator>>=(const M128 &a) { return *this = (Iu32vec4) _mm_srl_epi32(vec,a); }
|
||
|
Iu32vec4& operator>>=(int count) { return *this = (Iu32vec4) _mm_srli_epi32(vec,count); }
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator<< (std::ostream &os, const Iu32vec4 &a)
|
||
|
{
|
||
|
os << "[3]:" << _MM_4UDW(3,a)
|
||
|
<< " [2]:" << _MM_4UDW(2,a)
|
||
|
<< " [1]:" << _MM_4UDW(1,a)
|
||
|
<< " [0]:" << _MM_4UDW(0,a);
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const unsigned int& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 4); /* Only 4 elements to access */
|
||
|
return _MM_4UDW(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access and Assignment for Debug */
|
||
|
unsigned int& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 4); /* Only 4 elements to access */
|
||
|
return _MM_4UDW(i,vec);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
inline I64vec2 operator*(const Iu32vec4 &a, const Iu32vec4 &b) { return _mm_mul_epu32(a,b); }
|
||
|
inline Iu32vec4 cmpeq(const Iu32vec4 &a, const Iu32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
|
||
|
inline Iu32vec4 cmpneq(const Iu32vec4 &a, const Iu32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b), get_mask128()); }
|
||
|
|
||
|
inline Iu32vec4 unpack_low(const Iu32vec4 &a, const Iu32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
|
||
|
inline Iu32vec4 unpack_high(const Iu32vec4 &a, const Iu32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
|
||
|
|
||
|
/* I16vec8 Class:
|
||
|
* 8 elements, each element either unsigned or signed short
|
||
|
*/
|
||
|
class I16vec8 : public M128
|
||
|
{
|
||
|
public:
|
||
|
I16vec8() { }
|
||
|
I16vec8(__m128i mm) : M128(mm) { }
|
||
|
I16vec8(short s7, short s6, short s5, short s4, short s3, short s2, short s1, short s0)
|
||
|
{
|
||
|
vec = _mm_set_epi16(s7, s6, s5, s4, s3, s2, s1, s0);
|
||
|
}
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
I16vec8& operator= (const M128 &a) { return *this = (I16vec8) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
I16vec8& operator&=(const M128 &a) { return *this = (I16vec8) _mm_and_si128(vec,a); }
|
||
|
I16vec8& operator|=(const M128 &a) { return *this = (I16vec8) _mm_or_si128(vec,a); }
|
||
|
I16vec8& operator^=(const M128 &a) { return *this = (I16vec8) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
I16vec8& operator +=(const I16vec8 &a) { return *this = (I16vec8) _mm_add_epi16(vec,a); }
|
||
|
I16vec8& operator -=(const I16vec8 &a) { return *this = (I16vec8) _mm_sub_epi16(vec,a); }
|
||
|
I16vec8& operator *=(const I16vec8 &a) { return *this = (I16vec8) _mm_mullo_epi16(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
I16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
|
||
|
I16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
|
||
|
I16vec8& operator<<=(const M128 &a) { return *this = (I16vec8)_mm_sll_epi16(vec,a); }
|
||
|
I16vec8& operator<<=(int count) { return *this = (I16vec8)_mm_slli_epi16(vec,count); }
|
||
|
|
||
|
};
|
||
|
|
||
|
|
||
|
inline I16vec8 operator*(const I16vec8 &a, const I16vec8 &b) { return _mm_mullo_epi16(a,b); }
|
||
|
|
||
|
inline I16vec8 cmpeq(const I16vec8 &a, const I16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
|
||
|
inline I16vec8 cmpneq(const I16vec8 &a, const I16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b), get_mask128()); }
|
||
|
|
||
|
inline I16vec8 unpack_low(const I16vec8 &a, const I16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
|
||
|
inline I16vec8 unpack_high(const I16vec8 &a, const I16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
|
||
|
|
||
|
/* Is16vec8 Class:
|
||
|
* 8 elements, each element signed short
|
||
|
*/
|
||
|
class Is16vec8 : public I16vec8
|
||
|
{
|
||
|
public:
|
||
|
Is16vec8() { }
|
||
|
Is16vec8(__m128i mm) : I16vec8(mm) { }
|
||
|
Is16vec8(signed short s7, signed short s6, signed short s5,
|
||
|
signed short s4, signed short s3, signed short s2,
|
||
|
signed short s1, signed short s0)
|
||
|
: I16vec8(s7, s6, s5, s4, s3, s2, s1, s0) { }
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Is16vec8& operator= (const M128 &a) { return *this = (Is16vec8) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
Is16vec8& operator&=(const M128 &a) { return *this = (Is16vec8) _mm_and_si128(vec,a); }
|
||
|
Is16vec8& operator|=(const M128 &a) { return *this = (Is16vec8) _mm_or_si128(vec,a); }
|
||
|
Is16vec8& operator^=(const M128 &a) { return *this = (Is16vec8) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Is16vec8& operator +=(const I16vec8 &a) { return *this = (Is16vec8) _mm_add_epi16(vec,a); }
|
||
|
Is16vec8& operator -=(const I16vec8 &a) { return *this = (Is16vec8) _mm_sub_epi16(vec,a); }
|
||
|
Is16vec8& operator *=(const I16vec8 &a) { return *this = (Is16vec8) _mm_mullo_epi16(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
Is16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
|
||
|
Is16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
|
||
|
Is16vec8& operator<<=(const M128 &a) { return *this = (Is16vec8)_mm_sll_epi16(vec,a); }
|
||
|
Is16vec8& operator<<=(int count) { return *this = (Is16vec8)_mm_slli_epi16(vec,count); }
|
||
|
/* Shift Arithmetic Operators */
|
||
|
Is16vec8 operator>>(const M128 &a) { return _mm_sra_epi16(vec,a); }
|
||
|
Is16vec8 operator>>(int count) { return _mm_srai_epi16(vec,count); }
|
||
|
Is16vec8& operator>>=(const M128 &a) { return *this = (Is16vec8)_mm_sra_epi16(vec,a); }
|
||
|
Is16vec8& operator>>=(int count) { return *this = (Is16vec8)_mm_srai_epi16(vec,count); }
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator<< (std::ostream &os, const Is16vec8 &a)
|
||
|
{
|
||
|
os << "[7]:" << _MM_8W(7,a)
|
||
|
<< " [6]:" << _MM_8W(6,a)
|
||
|
<< " [5]:" << _MM_8W(5,a)
|
||
|
<< " [4]:" << _MM_8W(4,a)
|
||
|
<< " [3]:" << _MM_8W(3,a)
|
||
|
<< " [2]:" << _MM_8W(2,a)
|
||
|
<< " [1]:" << _MM_8W(1,a)
|
||
|
<< " [0]:" << _MM_8W(0,a);
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const signed short& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 8); /* Only 8 elements to access */
|
||
|
return _MM_8W(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access and Assignment for Debug */
|
||
|
signed short& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 8); /* Only 8 elements to access */
|
||
|
return _MM_8W(i,vec);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
inline Is16vec8 operator*(const Is16vec8 &a, const Is16vec8 &b) { return _mm_mullo_epi16(a,b); }
|
||
|
|
||
|
|
||
|
/* Additional Is16vec8 functions: compares, unpacks, sat add/sub */
|
||
|
inline Is16vec8 cmpeq(const Is16vec8 &a, const Is16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
|
||
|
inline Is16vec8 cmpneq(const Is16vec8 &a, const Is16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b), get_mask128()); }
|
||
|
inline Is16vec8 cmpgt(const Is16vec8 &a, const Is16vec8 &b) { return _mm_cmpgt_epi16(a,b); }
|
||
|
inline Is16vec8 cmplt(const Is16vec8 &a, const Is16vec8 &b) { return _mm_cmpgt_epi16(b,a); }
|
||
|
|
||
|
inline Is16vec8 unpack_low(const Is16vec8 &a, const Is16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
|
||
|
inline Is16vec8 unpack_high(const Is16vec8 &a, const Is16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
|
||
|
|
||
|
inline Is16vec8 mul_high(const Is16vec8 &a, const Is16vec8 &b) { return _mm_mulhi_epi16(a,b); }
|
||
|
inline Is32vec4 mul_add(const Is16vec8 &a, const Is16vec8 &b) { return _mm_madd_epi16(a,b);}
|
||
|
|
||
|
inline Is16vec8 sat_add(const Is16vec8 &a, const Is16vec8 &b) { return _mm_adds_epi16(a,b); }
|
||
|
inline Is16vec8 sat_sub(const Is16vec8 &a, const Is16vec8 &b) { return _mm_subs_epi16(a,b); }
|
||
|
|
||
|
inline Is16vec8 simd_max(const Is16vec8 &a, const Is16vec8 &b) { return _mm_max_epi16(a,b); }
|
||
|
inline Is16vec8 simd_min(const Is16vec8 &a, const Is16vec8 &b) { return _mm_min_epi16(a,b); }
|
||
|
|
||
|
|
||
|
/* Iu16vec8 Class:
|
||
|
* 8 elements, each element unsigned short
|
||
|
*/
|
||
|
class Iu16vec8 : public I16vec8
|
||
|
{
|
||
|
public:
|
||
|
Iu16vec8() { }
|
||
|
Iu16vec8(__m128i mm) : I16vec8(mm) { }
|
||
|
Iu16vec8(unsigned short s7, unsigned short s6, unsigned short s5,
|
||
|
unsigned short s4, unsigned short s3, unsigned short s2,
|
||
|
unsigned short s1, unsigned short s0)
|
||
|
: I16vec8(s7, s6, s5, s4, s3, s2, s1, s0) { }
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Iu16vec8& operator= (const M128 &a) { return *this = (Iu16vec8) a; }
|
||
|
/* Logical Assignment Operators */
|
||
|
Iu16vec8& operator&=(const M128 &a) { return *this = (Iu16vec8) _mm_and_si128(vec,a); }
|
||
|
Iu16vec8& operator|=(const M128 &a) { return *this = (Iu16vec8) _mm_or_si128(vec,a); }
|
||
|
Iu16vec8& operator^=(const M128 &a) { return *this = (Iu16vec8) _mm_xor_si128(vec,a); }
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Iu16vec8& operator +=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_add_epi16(vec,a); }
|
||
|
Iu16vec8& operator -=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_sub_epi16(vec,a); }
|
||
|
Iu16vec8& operator *=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_mullo_epi16(vec,a); }
|
||
|
|
||
|
/* Shift Logical Operators */
|
||
|
Iu16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
|
||
|
Iu16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
|
||
|
Iu16vec8& operator<<=(const M128 &a) { return *this = (Iu16vec8)_mm_sll_epi16(vec,a); }
|
||
|
Iu16vec8& operator<<=(int count) { return *this = (Iu16vec8)_mm_slli_epi16(vec,count); }
|
||
|
Iu16vec8 operator>>(const M128 &a) { return _mm_srl_epi16(vec,a); }
|
||
|
Iu16vec8 operator>>(int count) { return _mm_srli_epi16(vec,count); }
|
||
|
Iu16vec8& operator>>=(const M128 &a) { return *this = (Iu16vec8) _mm_srl_epi16(vec,a); }
|
||
|
Iu16vec8& operator>>=(int count) { return *this = (Iu16vec8) _mm_srli_epi16(vec,count); }
|
||
|
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator << (std::ostream &os, const Iu16vec8 &a)
|
||
|
{
|
||
|
os << "[7]:" << (unsigned short)(_MM_8UW(7,a))
|
||
|
<< " [6]:" << (unsigned short)(_MM_8UW(6,a))
|
||
|
<< " [5]:" << (unsigned short)(_MM_8UW(5,a))
|
||
|
<< " [4]:" << (unsigned short)(_MM_8UW(4,a))
|
||
|
<< " [3]:" << (unsigned short)(_MM_8UW(3,a))
|
||
|
<< " [2]:" << (unsigned short)(_MM_8UW(2,a))
|
||
|
<< " [1]:" << (unsigned short)(_MM_8UW(1,a))
|
||
|
<< " [0]:" << (unsigned short)(_MM_8UW(0,a));
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const unsigned short& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 8); /* Only 8 elements to access */
|
||
|
return _MM_8UW(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access for Debug */
|
||
|
unsigned short& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 8); /* Only 8 elements to access */
|
||
|
return _MM_8UW(i,vec);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
inline Iu16vec8 operator*(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_mullo_epi16(a,b); }
|
||
|
|
||
|
/* Additional Iu16vec8 functions: cmpeq,cmpneq, unpacks, sat add/sub */
|
||
|
inline Iu16vec8 cmpeq(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
|
||
|
inline Iu16vec8 cmpneq(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b), get_mask128()); }
|
||
|
|
||
|
inline Iu16vec8 unpack_low(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
|
||
|
inline Iu16vec8 unpack_high(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
|
||
|
|
||
|
inline Iu16vec8 sat_add(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_adds_epu16(a,b); }
|
||
|
inline Iu16vec8 sat_sub(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_subs_epu16(a,b); }
|
||
|
|
||
|
inline Iu16vec8 simd_avg(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_avg_epu16(a,b); }
|
||
|
inline I16vec8 mul_high(const Iu16vec8 &a, const Iu16vec8 &b) { return _mm_mulhi_epu16(a,b); }
|
||
|
|
||
|
/* I8vec16 Class:
|
||
|
* 16 elements, each element either unsigned or signed char
|
||
|
*/
|
||
|
class I8vec16 : public M128
|
||
|
{
|
||
|
public:
|
||
|
I8vec16() { }
|
||
|
I8vec16(__m128i mm) : M128(mm) { }
|
||
|
I8vec16(char s15, char s14, char s13, char s12, char s11, char s10,
|
||
|
char s9, char s8, char s7, char s6, char s5, char s4,
|
||
|
char s3, char s2, char s1, char s0)
|
||
|
{
|
||
|
vec = _mm_set_epi8(s15, s14, s13, s12, s11, s10, s9, s8, s7, s6, s5, s4, s3, s2, s1, s0);
|
||
|
}
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
I8vec16& operator= (const M128 &a) { return *this = (I8vec16) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
I8vec16& operator&=(const M128 &a) { return *this = (I8vec16) _mm_and_si128(vec,a); }
|
||
|
I8vec16& operator|=(const M128 &a) { return *this = (I8vec16) _mm_or_si128(vec,a); }
|
||
|
I8vec16& operator^=(const M128 &a) { return *this = (I8vec16) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
I8vec16& operator +=(const I8vec16 &a) { return *this = (I8vec16) _mm_add_epi8(vec,a); }
|
||
|
I8vec16& operator -=(const I8vec16 &a) { return *this = (I8vec16) _mm_sub_epi8(vec,a); }
|
||
|
|
||
|
};
|
||
|
|
||
|
inline I8vec16 cmpeq(const I8vec16 &a, const I8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
|
||
|
inline I8vec16 cmpneq(const I8vec16 &a, const I8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b), get_mask128()); }
|
||
|
|
||
|
inline I8vec16 unpack_low(const I8vec16 &a, const I8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
|
||
|
inline I8vec16 unpack_high(const I8vec16 &a, const I8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
|
||
|
|
||
|
/* Is8vec16 Class:
|
||
|
* 16 elements, each element a signed char
|
||
|
*/
|
||
|
class Is8vec16 : public I8vec16
|
||
|
{
|
||
|
public:
|
||
|
Is8vec16() { }
|
||
|
Is8vec16(__m128i mm) : I8vec16(mm) { }
|
||
|
Is8vec16(char s15, char s14, char s13, char s12, char s11, char s10,
|
||
|
char s9, char s8, char s7, char s6, char s5, char s4,
|
||
|
char s3, char s2, char s1, char s0)
|
||
|
: I8vec16(s15, s14, s13, s12, s11, s10, s9, s8,
|
||
|
s7, s6, s5, s4, s3, s2, s1, s0) { }
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Is8vec16& operator= (const M128 &a) { return *this = (Is8vec16) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
Is8vec16& operator&=(const M128 &a) { return *this = (Is8vec16) _mm_and_si128(vec,a); }
|
||
|
Is8vec16& operator|=(const M128 &a) { return *this = (Is8vec16) _mm_or_si128(vec,a); }
|
||
|
Is8vec16& operator^=(const M128 &a) { return *this = (Is8vec16) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Is8vec16& operator +=(const I8vec16 &a) { return *this = (Is8vec16) _mm_add_epi8(vec,a); }
|
||
|
Is8vec16& operator -=(const I8vec16 &a) { return *this = (Is8vec16) _mm_sub_epi8(vec,a); }
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator << (std::ostream &os, const Is8vec16 &a)
|
||
|
{
|
||
|
os << "[15]:" << short(_MM_16B(15,a))
|
||
|
<< " [14]:" << short(_MM_16B(14,a))
|
||
|
<< " [13]:" << short(_MM_16B(13,a))
|
||
|
<< " [12]:" << short(_MM_16B(12,a))
|
||
|
<< " [11]:" << short(_MM_16B(11,a))
|
||
|
<< " [10]:" << short(_MM_16B(10,a))
|
||
|
<< " [9]:" << short(_MM_16B(9,a))
|
||
|
<< " [8]:" << short(_MM_16B(8,a))
|
||
|
<< " [7]:" << short(_MM_16B(7,a))
|
||
|
<< " [6]:" << short(_MM_16B(6,a))
|
||
|
<< " [5]:" << short(_MM_16B(5,a))
|
||
|
<< " [4]:" << short(_MM_16B(4,a))
|
||
|
<< " [3]:" << short(_MM_16B(3,a))
|
||
|
<< " [2]:" << short(_MM_16B(2,a))
|
||
|
<< " [1]:" << short(_MM_16B(1,a))
|
||
|
<< " [0]:" << short(_MM_16B(0,a));
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const signed char& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 16); /* Only 16 elements to access */
|
||
|
return _MM_16B(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access for Debug */
|
||
|
signed char& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 16); /* Only 16 elements to access */
|
||
|
return _MM_16B(i,vec);
|
||
|
}
|
||
|
|
||
|
};
|
||
|
|
||
|
inline Is8vec16 cmpeq(const Is8vec16 &a, const Is8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
|
||
|
inline Is8vec16 cmpneq(const Is8vec16 &a, const Is8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b), get_mask128()); }
|
||
|
inline Is8vec16 cmpgt(const Is8vec16 &a, const Is8vec16 &b) { return _mm_cmpgt_epi8(a,b); }
|
||
|
inline Is8vec16 cmplt(const Is8vec16 &a, const Is8vec16 &b) { return _mm_cmplt_epi8(a,b); }
|
||
|
|
||
|
inline Is8vec16 unpack_low(const Is8vec16 &a, const Is8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
|
||
|
inline Is8vec16 unpack_high(const Is8vec16 &a, const Is8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
|
||
|
|
||
|
inline Is8vec16 sat_add(const Is8vec16 &a, const Is8vec16 &b) { return _mm_adds_epi8(a,b); }
|
||
|
inline Is8vec16 sat_sub(const Is8vec16 &a, const Is8vec16 &b) { return _mm_subs_epi8(a,b); }
|
||
|
|
||
|
/* Iu8vec16 Class:
|
||
|
* 16 elements, each element a unsigned char
|
||
|
*/
|
||
|
class Iu8vec16 : public I8vec16
|
||
|
{
|
||
|
public:
|
||
|
Iu8vec16() { }
|
||
|
Iu8vec16(__m128i mm) : I8vec16(mm) { }
|
||
|
Iu8vec16(unsigned char u15, unsigned char u14, unsigned char u13,
|
||
|
unsigned char u12, unsigned char u11, unsigned char u10,
|
||
|
unsigned char u9, unsigned char u8, unsigned char u7,
|
||
|
unsigned char u6, unsigned char u5, unsigned char u4,
|
||
|
unsigned char u3, unsigned char u2, unsigned char u1,
|
||
|
unsigned char u0)
|
||
|
: I8vec16(u15, u14, u13, u12, u11, u10, u9, u8,
|
||
|
u7, u6, u5, u4, u3, u2, u1, u0) { }
|
||
|
|
||
|
/* Assignment Operator */
|
||
|
Iu8vec16& operator= (const M128 &a) { return *this = (Iu8vec16) a; }
|
||
|
|
||
|
/* Logical Assignment Operators */
|
||
|
Iu8vec16& operator&=(const M128 &a) { return *this = (Iu8vec16) _mm_and_si128(vec,a); }
|
||
|
Iu8vec16& operator|=(const M128 &a) { return *this = (Iu8vec16) _mm_or_si128(vec,a); }
|
||
|
Iu8vec16& operator^=(const M128 &a) { return *this = (Iu8vec16) _mm_xor_si128(vec,a); }
|
||
|
|
||
|
/* Addition & Subtraction Assignment Operators */
|
||
|
Iu8vec16& operator +=(const I8vec16 &a) { return *this = (Iu8vec16) _mm_add_epi8(vec,a); }
|
||
|
Iu8vec16& operator -=(const I8vec16 &a) { return *this = (Iu8vec16) _mm_sub_epi8(vec,a); }
|
||
|
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output for Debug */
|
||
|
friend std::ostream& operator << (std::ostream &os, const Iu8vec16 &a)
|
||
|
{
|
||
|
os << "[15]:" << (unsigned char)(_MM_16UB(15,a))
|
||
|
<< " [14]:" << (unsigned char)(_MM_16UB(14,a))
|
||
|
<< " [13]:" << (unsigned char)(_MM_16UB(13,a))
|
||
|
<< " [12]:" << (unsigned char)(_MM_16UB(12,a))
|
||
|
<< " [11]:" << (unsigned char)(_MM_16UB(11,a))
|
||
|
<< " [10]:" << (unsigned char)(_MM_16UB(10,a))
|
||
|
<< " [9]:" << (unsigned char)(_MM_16UB(9,a))
|
||
|
<< " [8]:" << (unsigned char)(_MM_16UB(8,a))
|
||
|
<< " [7]:" << (unsigned char)(_MM_16UB(7,a))
|
||
|
<< " [6]:" << (unsigned char)(_MM_16UB(6,a))
|
||
|
<< " [5]:" << (unsigned char)(_MM_16UB(5,a))
|
||
|
<< " [4]:" << (unsigned char)(_MM_16UB(4,a))
|
||
|
<< " [3]:" << (unsigned char)(_MM_16UB(3,a))
|
||
|
<< " [2]:" << (unsigned char)(_MM_16UB(2,a))
|
||
|
<< " [1]:" << (unsigned char)(_MM_16UB(1,a))
|
||
|
<< " [0]:" << (unsigned char)(_MM_16UB(0,a));
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access for Debug, No data modified */
|
||
|
const unsigned char& operator[](int i)const
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 16); /* Only 16 elements to access */
|
||
|
return _MM_16UB(i,vec);
|
||
|
}
|
||
|
|
||
|
/* Element Access for Debug */
|
||
|
unsigned char& operator[](int i)
|
||
|
{
|
||
|
_VEC_ASSERT(static_cast<unsigned int>(i) < 16); /* Only 16 elements to access */
|
||
|
return _MM_16UB(i,vec);
|
||
|
}
|
||
|
|
||
|
};
|
||
|
|
||
|
inline Iu8vec16 cmpeq(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
|
||
|
inline Iu8vec16 cmpneq(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b), get_mask128()); }
|
||
|
|
||
|
inline Iu8vec16 unpack_low(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
|
||
|
inline Iu8vec16 unpack_high(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
|
||
|
|
||
|
inline Iu8vec16 sat_add(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_adds_epu8(a,b); }
|
||
|
inline Iu8vec16 sat_sub(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_subs_epu8(a,b); }
|
||
|
|
||
|
inline I64vec2 sum_abs(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_sad_epu8(a,b); }
|
||
|
|
||
|
inline Iu8vec16 simd_avg(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_avg_epu8(a,b); }
|
||
|
inline Iu8vec16 simd_max(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_max_epu8(a,b); }
|
||
|
inline Iu8vec16 simd_min(const Iu8vec16 &a, const Iu8vec16 &b) { return _mm_min_epu8(a,b); }
|
||
|
|
||
|
/* Pack & Saturates */
|
||
|
|
||
|
inline Is16vec8 pack_sat(const Is32vec4 &a, const Is32vec4 &b) { return _mm_packs_epi32(a,b); }
|
||
|
inline Is8vec16 pack_sat(const Is16vec8 &a, const Is16vec8 &b) { return _mm_packs_epi16(a,b); }
|
||
|
inline Iu8vec16 packu_sat(const Is16vec8 &a, const Is16vec8 &b) { return _mm_packus_epi16(a,b);}
|
||
|
|
||
|
/********************************* Logicals ****************************************/
|
||
|
#define IVEC128_LOGICALS(vect,element) \
|
||
|
inline I##vect##vec##element operator& (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_and_si128( a,b); } \
|
||
|
inline I##vect##vec##element operator| (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_or_si128( a,b); } \
|
||
|
inline I##vect##vec##element operator^ (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_xor_si128( a,b); } \
|
||
|
inline I##vect##vec##element andnot (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_andnot_si128( a,b); }
|
||
|
|
||
|
IVEC128_LOGICALS(8,16)
|
||
|
IVEC128_LOGICALS(u8,16)
|
||
|
IVEC128_LOGICALS(s8,16)
|
||
|
IVEC128_LOGICALS(16,8)
|
||
|
IVEC128_LOGICALS(u16,8)
|
||
|
IVEC128_LOGICALS(s16,8)
|
||
|
IVEC128_LOGICALS(32,4)
|
||
|
IVEC128_LOGICALS(u32,4)
|
||
|
IVEC128_LOGICALS(s32,4)
|
||
|
IVEC128_LOGICALS(64,2)
|
||
|
IVEC128_LOGICALS(128,1)
|
||
|
#undef IVEC128_LOGICALS
|
||
|
|
||
|
/********************************* Add & Sub ****************************************/
|
||
|
#define IVEC128_ADD_SUB(vect,element,opsize) \
|
||
|
inline I##vect##vec##element operator+ (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_add_##opsize( a,b); } \
|
||
|
inline I##vect##vec##element operator- (const I##vect##vec##element &a, const I##vect##vec##element &b) \
|
||
|
{ return _mm_sub_##opsize( a,b); }
|
||
|
|
||
|
IVEC128_ADD_SUB(8,16, epi8)
|
||
|
IVEC128_ADD_SUB(u8,16, epi8)
|
||
|
IVEC128_ADD_SUB(s8,16, epi8)
|
||
|
IVEC128_ADD_SUB(16,8, epi16)
|
||
|
IVEC128_ADD_SUB(u16,8, epi16)
|
||
|
IVEC128_ADD_SUB(s16,8, epi16)
|
||
|
IVEC128_ADD_SUB(32,4, epi32)
|
||
|
IVEC128_ADD_SUB(u32,4, epi32)
|
||
|
IVEC128_ADD_SUB(s32,4, epi32)
|
||
|
IVEC128_ADD_SUB(64,2, epi64)
|
||
|
#undef IVEC128_ADD_SUB
|
||
|
|
||
|
/************************* Conditional Select ********************************
|
||
|
* version of: retval = (a OP b)? c : d; *
|
||
|
* Where OP is one of the possible comparision operators. *
|
||
|
* Example: r = select_eq(a,b,c,d); *
|
||
|
* if "member at position x of the vector a" == *
|
||
|
* "member at position x of vector b" *
|
||
|
* assign the corresponding member in r from c, else assign from d. *
|
||
|
************************* Conditional Select ********************************/
|
||
|
|
||
|
#define IVEC128_SELECT(vect12,vect34,element,selop) \
|
||
|
inline I##vect34##vec##element select_##selop ( \
|
||
|
const I##vect12##vec##element &a, \
|
||
|
const I##vect12##vec##element &b, \
|
||
|
const I##vect34##vec##element &c, \
|
||
|
const I##vect34##vec##element &d) \
|
||
|
{ \
|
||
|
I##vect12##vec##element mask = cmp##selop(a,b); \
|
||
|
return ( I##vect34##vec##element (mask & c ) | \
|
||
|
I##vect34##vec##element ((_mm_andnot_si128(mask, d )))); \
|
||
|
}
|
||
|
|
||
|
IVEC128_SELECT(8,s8,16,eq)
|
||
|
IVEC128_SELECT(8,u8,16,eq)
|
||
|
IVEC128_SELECT(8,8,16,eq)
|
||
|
IVEC128_SELECT(8,s8,16,neq)
|
||
|
IVEC128_SELECT(8,u8,16,neq)
|
||
|
IVEC128_SELECT(8,8,16,neq)
|
||
|
|
||
|
IVEC128_SELECT(16,s16,8,eq)
|
||
|
IVEC128_SELECT(16,u16,8,eq)
|
||
|
IVEC128_SELECT(16,16,8,eq)
|
||
|
IVEC128_SELECT(16,s16,8,neq)
|
||
|
IVEC128_SELECT(16,u16,8,neq)
|
||
|
IVEC128_SELECT(16,16,8,neq)
|
||
|
|
||
|
IVEC128_SELECT(32,s32,4,eq)
|
||
|
IVEC128_SELECT(32,u32,4,eq)
|
||
|
IVEC128_SELECT(32,32,4,eq)
|
||
|
IVEC128_SELECT(32,s32,4,neq)
|
||
|
IVEC128_SELECT(32,u32,4,neq)
|
||
|
IVEC128_SELECT(32,32,4,neq)
|
||
|
|
||
|
IVEC128_SELECT(s8,s8,16,gt)
|
||
|
IVEC128_SELECT(s8,u8,16,gt)
|
||
|
IVEC128_SELECT(s8,8,16,gt)
|
||
|
IVEC128_SELECT(s8,s8,16,lt)
|
||
|
IVEC128_SELECT(s8,u8,16,lt)
|
||
|
IVEC128_SELECT(s8,8,16,lt)
|
||
|
|
||
|
IVEC128_SELECT(s16,s16,8,gt)
|
||
|
IVEC128_SELECT(s16,u16,8,gt)
|
||
|
IVEC128_SELECT(s16,16,8,gt)
|
||
|
IVEC128_SELECT(s16,s16,8,lt)
|
||
|
IVEC128_SELECT(s16,u16,8,lt)
|
||
|
IVEC128_SELECT(s16,16,8,lt)
|
||
|
|
||
|
|
||
|
#undef IVEC128_SELECT
|
||
|
|
||
|
|
||
|
class F64vec2
|
||
|
{
|
||
|
protected:
|
||
|
__m128d vec;
|
||
|
public:
|
||
|
|
||
|
/* Constructors: __m128d, 2 doubles */
|
||
|
F64vec2() {}
|
||
|
|
||
|
/* initialize 2 DP FP with __m128d data type */
|
||
|
F64vec2(__m128d m) { vec = m;}
|
||
|
|
||
|
/* initialize 2 DP FPs with 2 doubles */
|
||
|
F64vec2(double d1, double d0) { vec= _mm_set_pd(d1,d0); }
|
||
|
|
||
|
/* Explicitly initialize each of 2 DP FPs with same double */
|
||
|
EXPLICIT F64vec2(double d) { vec = _mm_set1_pd(d); }
|
||
|
|
||
|
/* Conversion functions */
|
||
|
operator __m128d() const { return vec; } /* Convert to __m128d */
|
||
|
|
||
|
/* Logical Operators */
|
||
|
friend F64vec2 operator &(const F64vec2 &a, const F64vec2 &b) { return _mm_and_pd(a,b); }
|
||
|
friend F64vec2 operator |(const F64vec2 &a, const F64vec2 &b) { return _mm_or_pd(a,b); }
|
||
|
friend F64vec2 operator ^(const F64vec2 &a, const F64vec2 &b) { return _mm_xor_pd(a,b); }
|
||
|
|
||
|
/* Arithmetic Operators */
|
||
|
friend F64vec2 operator +(const F64vec2 &a, const F64vec2 &b) { return _mm_add_pd(a,b); }
|
||
|
friend F64vec2 operator -(const F64vec2 &a, const F64vec2 &b) { return _mm_sub_pd(a,b); }
|
||
|
friend F64vec2 operator *(const F64vec2 &a, const F64vec2 &b) { return _mm_mul_pd(a,b); }
|
||
|
friend F64vec2 operator /(const F64vec2 &a, const F64vec2 &b) { return _mm_div_pd(a,b); }
|
||
|
|
||
|
F64vec2& operator +=(const F64vec2 &a) { return *this = _mm_add_pd(vec,a); }
|
||
|
F64vec2& operator -=(const F64vec2 &a) { return *this = _mm_sub_pd(vec,a); }
|
||
|
F64vec2& operator *=(const F64vec2 &a) { return *this = _mm_mul_pd(vec,a); }
|
||
|
F64vec2& operator /=(const F64vec2 &a) { return *this = _mm_div_pd(vec,a); }
|
||
|
F64vec2& operator &=(const F64vec2 &a) { return *this = _mm_and_pd(vec,a); }
|
||
|
F64vec2& operator |=(const F64vec2 &a) { return *this = _mm_or_pd(vec,a); }
|
||
|
F64vec2& operator ^=(const F64vec2 &a) { return *this = _mm_xor_pd(vec,a); }
|
||
|
|
||
|
/* Horizontal Add */
|
||
|
friend double add_horizontal(const F64vec2 &a)
|
||
|
{
|
||
|
F64vec2 ftemp = _mm_add_sd(a,_mm_shuffle_pd(a, a, 1));
|
||
|
return _mm_cvtsd_f64(ftemp);
|
||
|
}
|
||
|
|
||
|
/* And Not */
|
||
|
friend F64vec2 andnot(const F64vec2 &a, const F64vec2 &b) { return _mm_andnot_pd(a,b); }
|
||
|
|
||
|
/* Square Root */
|
||
|
friend F64vec2 sqrt(const F64vec2 &a) { return _mm_sqrt_pd(a); }
|
||
|
|
||
|
/* Compares: Mask is returned */
|
||
|
/* Macros expand to all compare intrinsics. Example:
|
||
|
friend F64vec2 cmpeq(const F64vec2 &a, const F64vec2 &b)
|
||
|
{ return _mm_cmpeq_ps(a,b);} */
|
||
|
#define F64vec2_COMP(op) \
|
||
|
friend F64vec2 cmp##op (const F64vec2 &a, const F64vec2 &b) { return _mm_cmp##op##_pd(a,b); }
|
||
|
F64vec2_COMP(eq) /* expanded to cmpeq(a,b) */
|
||
|
F64vec2_COMP(lt) /* expanded to cmplt(a,b) */
|
||
|
F64vec2_COMP(le) /* expanded to cmple(a,b) */
|
||
|
F64vec2_COMP(gt) /* expanded to cmpgt(a,b) */
|
||
|
F64vec2_COMP(ge) /* expanded to cmpge(a,b) */
|
||
|
F64vec2_COMP(ngt) /* expanded to cmpngt(a,b) */
|
||
|
F64vec2_COMP(nge) /* expanded to cmpnge(a,b) */
|
||
|
F64vec2_COMP(neq) /* expanded to cmpneq(a,b) */
|
||
|
F64vec2_COMP(nlt) /* expanded to cmpnlt(a,b) */
|
||
|
F64vec2_COMP(nle) /* expanded to cmpnle(a,b) */
|
||
|
#undef F64vec2_COMP
|
||
|
|
||
|
/* Min and Max */
|
||
|
friend F64vec2 simd_min(const F64vec2 &a, const F64vec2 &b) { return _mm_min_pd(a,b); }
|
||
|
friend F64vec2 simd_max(const F64vec2 &a, const F64vec2 &b) { return _mm_max_pd(a,b); }
|
||
|
|
||
|
/* Absolute value */
|
||
|
friend F64vec2 abs(const F64vec2 &a)
|
||
|
{
|
||
|
return _mm_and_pd(a, _f64vec2_abs_mask);
|
||
|
}
|
||
|
|
||
|
/* Compare lower DP FP values */
|
||
|
#define F64vec2_COMI(op) \
|
||
|
friend int comi##op (const F64vec2 &a, const F64vec2 &b) { return _mm_comi##op##_sd(a,b); }
|
||
|
F64vec2_COMI(eq) /* expanded to comieq(a,b) */
|
||
|
F64vec2_COMI(lt) /* expanded to comilt(a,b) */
|
||
|
F64vec2_COMI(le) /* expanded to comile(a,b) */
|
||
|
F64vec2_COMI(gt) /* expanded to comigt(a,b) */
|
||
|
F64vec2_COMI(ge) /* expanded to comige(a,b) */
|
||
|
F64vec2_COMI(neq) /* expanded to comineq(a,b) */
|
||
|
#undef F64vec2_COMI
|
||
|
|
||
|
/* Compare lower DP FP values */
|
||
|
#define F64vec2_UCOMI(op) \
|
||
|
friend int ucomi##op (const F64vec2 &a, const F64vec2 &b) { return _mm_ucomi##op##_sd(a,b); }
|
||
|
F64vec2_UCOMI(eq) /* expanded to ucomieq(a,b) */
|
||
|
F64vec2_UCOMI(lt) /* expanded to ucomilt(a,b) */
|
||
|
F64vec2_UCOMI(le) /* expanded to ucomile(a,b) */
|
||
|
F64vec2_UCOMI(gt) /* expanded to ucomigt(a,b) */
|
||
|
F64vec2_UCOMI(ge) /* expanded to ucomige(a,b) */
|
||
|
F64vec2_UCOMI(neq) /* expanded to ucomineq(a,b) */
|
||
|
#undef F64vec2_UCOMI
|
||
|
|
||
|
/* Debug Features */
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output */
|
||
|
friend std::ostream & operator<<(std::ostream & os, const F64vec2 &a)
|
||
|
{
|
||
|
/* To use: cout << "Elements of F64vec2 fvec are: " << fvec; */
|
||
|
double *dp = (double*)&a;
|
||
|
os << "[1]:" << *(dp+1)
|
||
|
<< " [0]:" << *dp;
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
/* Element Access Only, no modifications to elements*/
|
||
|
const double& operator[](int i) const
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 1)); /* User should only access elements 0-1 */
|
||
|
double *dp = (double*)&vec;
|
||
|
return *(dp+i);
|
||
|
}
|
||
|
/* Element Access and Modification*/
|
||
|
double& operator[](int i)
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 1)); /* User should only access elements 0-1 */
|
||
|
double *dp = (double*)&vec;
|
||
|
return *(dp+i);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
/* Miscellaneous */
|
||
|
|
||
|
/* Interleave low order data elements of a and b into destination */
|
||
|
inline F64vec2 unpack_low(const F64vec2 &a, const F64vec2 &b)
|
||
|
{ return _mm_unpacklo_pd(a, b); }
|
||
|
|
||
|
/* Interleave high order data elements of a and b into target */
|
||
|
inline F64vec2 unpack_high(const F64vec2 &a, const F64vec2 &b)
|
||
|
{ return _mm_unpackhi_pd(a, b); }
|
||
|
|
||
|
/* Move Mask to Integer returns 4 bit mask formed of most significant bits of a */
|
||
|
inline int move_mask(const F64vec2 &a)
|
||
|
{ return _mm_movemask_pd(a);}
|
||
|
|
||
|
/* Data Motion Functions */
|
||
|
|
||
|
/* Load Unaligned loadu_pd: Unaligned */
|
||
|
inline void loadu(F64vec2 &a, double *p)
|
||
|
{ a = _mm_loadu_pd(p); }
|
||
|
|
||
|
/* Store Temporal storeu_pd: Unaligned */
|
||
|
inline void storeu(double *p, const F64vec2 &a)
|
||
|
{ _mm_storeu_pd(p, a); }
|
||
|
|
||
|
/* Cacheability Support */
|
||
|
|
||
|
/* Non-Temporal Store */
|
||
|
inline void store_nta(double *p, F64vec2 &a)
|
||
|
{ _mm_stream_pd(p,a);}
|
||
|
|
||
|
#define F64vec2_SELECT(op) \
|
||
|
inline F64vec2 select_##op (const F64vec2 &a, const F64vec2 &b, const F64vec2 &c, const F64vec2 &d) \
|
||
|
{ \
|
||
|
F64vec2 mask = _mm_cmp##op##_pd(a,b); \
|
||
|
return( (mask & c) | F64vec2((_mm_andnot_pd(mask,d)))); \
|
||
|
}
|
||
|
F64vec2_SELECT(eq) /* generates select_eq(a,b) */
|
||
|
F64vec2_SELECT(lt) /* generates select_lt(a,b) */
|
||
|
F64vec2_SELECT(le) /* generates select_le(a,b) */
|
||
|
F64vec2_SELECT(gt) /* generates select_gt(a,b) */
|
||
|
F64vec2_SELECT(ge) /* generates select_ge(a,b) */
|
||
|
F64vec2_SELECT(neq) /* generates select_neq(a,b) */
|
||
|
F64vec2_SELECT(nlt) /* generates select_nlt(a,b) */
|
||
|
F64vec2_SELECT(nle) /* generates select_nle(a,b) */
|
||
|
#undef F64vec2_SELECT
|
||
|
|
||
|
/* Convert the lower DP FP value of a to a 32 bit signed integer using Truncate*/
|
||
|
inline int F64vec2ToInt(const F64vec2 &a)
|
||
|
{
|
||
|
|
||
|
return _mm_cvttsd_si32(a);
|
||
|
|
||
|
}
|
||
|
|
||
|
/* Convert the 4 SP FP values of a to DP FP values */
|
||
|
inline F64vec2 F32vec4ToF64vec2(const F32vec4 &a)
|
||
|
{
|
||
|
return _mm_cvtps_pd(a);
|
||
|
}
|
||
|
|
||
|
/* Convert the 2 DP FP values of a to SP FP values */
|
||
|
inline F32vec4 F64vec2ToF32vec4(const F64vec2 &a)
|
||
|
{
|
||
|
return _mm_cvtpd_ps(a);
|
||
|
}
|
||
|
|
||
|
/* Convert the signed int in b to a DP FP value. Upper DP FP value in a passed through */
|
||
|
inline F64vec2 IntToF64vec2(const F64vec2 &a, int b)
|
||
|
{
|
||
|
return _mm_cvtsi32_sd(a,b);
|
||
|
}
|
||
|
|
||
|
#pragma pack(pop) /* 16-B aligned */
|
||
|
|
||
|
/******************************************************************************/
|
||
|
/************** Interface classes for Intel(R) AVX intrinsics *****************/
|
||
|
/******************************************************************************/
|
||
|
|
||
|
/*
|
||
|
* class F32vec8
|
||
|
*
|
||
|
* Represents 256-bit vector composed of 8 single precision floating point elements.
|
||
|
*/
|
||
|
class F32vec8
|
||
|
{
|
||
|
protected:
|
||
|
__m256 vec;
|
||
|
|
||
|
public:
|
||
|
|
||
|
/* Constructors: __m256, 8 floats, 1 float */
|
||
|
F32vec8() {}
|
||
|
|
||
|
/* initialize 8 SP FP with __m256 data type */
|
||
|
F32vec8(__m256 m) { vec = m; }
|
||
|
|
||
|
/* initialize 8 SP FPs with 8 floats */
|
||
|
F32vec8(float f7, float f6, float f5, float f4, float f3, float f2, float f1, float f0)
|
||
|
{
|
||
|
vec = _mm256_set_ps(f7,f6,f5,f4,f3,f2,f1,f0);
|
||
|
}
|
||
|
|
||
|
/* Explicitly initialize each of 8 SP FPs with same float */
|
||
|
EXPLICIT F32vec8(float f) { vec = _mm256_set1_ps(f); }
|
||
|
|
||
|
/* Explicitly initialize each of 8 SP FPs with same double */
|
||
|
EXPLICIT F32vec8(double d) { vec = _mm256_set1_ps((float) d); }
|
||
|
|
||
|
/* Assignment operations */
|
||
|
F32vec8& operator =(float f)
|
||
|
{
|
||
|
vec = _mm256_set1_ps(f);
|
||
|
return *this;
|
||
|
}
|
||
|
|
||
|
F32vec8& operator =(double d)
|
||
|
{
|
||
|
vec = _mm256_set1_ps((float) d);
|
||
|
return *this;
|
||
|
}
|
||
|
|
||
|
/* Conversion functions */
|
||
|
operator __m256() const { return vec; }
|
||
|
|
||
|
/* Logical Operators */
|
||
|
friend F32vec8 operator &(const F32vec8 &a, const F32vec8 &b) { return _mm256_and_ps(a,b); }
|
||
|
friend F32vec8 operator |(const F32vec8 &a, const F32vec8 &b) { return _mm256_or_ps(a,b); }
|
||
|
friend F32vec8 operator ^(const F32vec8 &a, const F32vec8 &b) { return _mm256_xor_ps(a,b); }
|
||
|
|
||
|
/* Arithmetic Operators */
|
||
|
friend F32vec8 operator +(const F32vec8 &a, const F32vec8 &b) { return _mm256_add_ps(a,b); }
|
||
|
friend F32vec8 operator -(const F32vec8 &a, const F32vec8 &b) { return _mm256_sub_ps(a,b); }
|
||
|
friend F32vec8 operator *(const F32vec8 &a, const F32vec8 &b) { return _mm256_mul_ps(a,b); }
|
||
|
friend F32vec8 operator /(const F32vec8 &a, const F32vec8 &b) { return _mm256_div_ps(a,b); }
|
||
|
|
||
|
F32vec8& operator +=(const F32vec8 &a) { return *this = _mm256_add_ps(vec,a); }
|
||
|
F32vec8& operator -=(const F32vec8 &a) { return *this = _mm256_sub_ps(vec,a); }
|
||
|
F32vec8& operator *=(const F32vec8 &a) { return *this = _mm256_mul_ps(vec,a); }
|
||
|
F32vec8& operator /=(const F32vec8 &a) { return *this = _mm256_div_ps(vec,a); }
|
||
|
F32vec8& operator &=(const F32vec8 &a) { return *this = _mm256_and_ps(vec,a); }
|
||
|
F32vec8& operator |=(const F32vec8 &a) { return *this = _mm256_or_ps(vec,a); }
|
||
|
F32vec8& operator ^=(const F32vec8 &a) { return *this = _mm256_xor_ps(vec,a); }
|
||
|
|
||
|
/* Horizontal Add */
|
||
|
friend float add_horizontal(const F32vec8 &a)
|
||
|
{
|
||
|
F32vec8 temp = _mm256_add_ps(a, _mm256_permute_ps(a, 0xee));
|
||
|
temp = _mm256_add_ps(temp, _mm256_movehdup_ps(temp));
|
||
|
return _mm_cvtss_f32(_mm_add_ss(_mm256_castps256_ps128(temp), _mm256_extractf128_ps(temp,1)));
|
||
|
}
|
||
|
|
||
|
/* And Not */
|
||
|
friend F32vec8 andnot(const F32vec8 &a, const F32vec8 &b) { return _mm256_andnot_ps(a,b); }
|
||
|
|
||
|
/* Square Root */
|
||
|
friend F32vec8 sqrt(const F32vec8 &a) { return _mm256_sqrt_ps(a); }
|
||
|
|
||
|
/* Reciprocal */
|
||
|
friend F32vec8 rcp(const F32vec8 &a) { return _mm256_rcp_ps(a); }
|
||
|
|
||
|
/* Reciprocal Square Root */
|
||
|
friend F32vec8 rsqrt(const F32vec8 &a) { return _mm256_rsqrt_ps(a); }
|
||
|
|
||
|
/*
|
||
|
* NewtonRaphson Reciprocal
|
||
|
* [2 * rcpps(x) - (x * rcpps(x) * rcpps(x))]
|
||
|
*/
|
||
|
friend F32vec8 rcp_nr(const F32vec8 &a)
|
||
|
{
|
||
|
F32vec8 Ra0 = _mm256_rcp_ps(a);
|
||
|
return _mm256_sub_ps(_mm256_add_ps(Ra0, Ra0), _mm256_mul_ps(_mm256_mul_ps(Ra0, a), Ra0));
|
||
|
}
|
||
|
|
||
|
/*
|
||
|
* NewtonRaphson Reciprocal Square Root
|
||
|
* 0.5 * rsqrtps * (3 - x * rsqrtps(x) * rsqrtps(x))
|
||
|
*/
|
||
|
friend F32vec8 rsqrt_nr(const F32vec8 &a)
|
||
|
{
|
||
|
#pragma warning(push)
|
||
|
#pragma warning(disable:4640)
|
||
|
static const F32vec8 fvecf0pt5(0.5f);
|
||
|
static const F32vec8 fvecf3pt0(3.0f);
|
||
|
#pragma warning(pop)
|
||
|
F32vec8 Ra0 = _mm256_rsqrt_ps(a);
|
||
|
return (fvecf0pt5 * Ra0) * (fvecf3pt0 - (a * Ra0) * Ra0);
|
||
|
|
||
|
}
|
||
|
|
||
|
/* Compares: Mask is returned */
|
||
|
friend F32vec8 cmp_eq(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_EQ_OQ); }
|
||
|
friend F32vec8 cmp_lt(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_LT_OS); }
|
||
|
friend F32vec8 cmp_le(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_LE_OS); }
|
||
|
friend F32vec8 cmp_gt(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_GT_OS); }
|
||
|
friend F32vec8 cmp_ge(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_GE_OS); }
|
||
|
friend F32vec8 cmp_neq(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_NEQ_UQ); }
|
||
|
friend F32vec8 cmp_nlt(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_NLT_US); }
|
||
|
friend F32vec8 cmp_nle(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_NLE_US); }
|
||
|
friend F32vec8 cmp_ngt(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_NGT_US); }
|
||
|
friend F32vec8 cmp_nge(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_cmp_ps(a, b, _CMP_NGE_US); }
|
||
|
|
||
|
/* Min and Max */
|
||
|
friend F32vec8 simd_min(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_min_ps(a,b); }
|
||
|
friend F32vec8 simd_max(const F32vec8 &a, const F32vec8 &b)
|
||
|
{ return _mm256_max_ps(a,b); }
|
||
|
|
||
|
/* Absolute value */
|
||
|
friend F32vec8 abs(const F32vec8 &a)
|
||
|
{
|
||
|
static const union
|
||
|
{
|
||
|
int i[8];
|
||
|
__m256 m;
|
||
|
} __f32vec8_abs_mask = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff,
|
||
|
0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff};
|
||
|
return _mm256_and_ps(a, __f32vec8_abs_mask.m);
|
||
|
}
|
||
|
|
||
|
/* Debug Features */
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output */
|
||
|
friend DVEC_STD ostream & operator<<(DVEC_STD ostream &os, const F32vec8 &a)
|
||
|
{
|
||
|
/* To use: cout << "Elements of F32vec8 fvec are: " << fvec; */
|
||
|
float *fp = (float*) &a;
|
||
|
os << "[7]:" << *(fp+7)
|
||
|
<< " [6]:" << *(fp+6)
|
||
|
<< " [5]:" << *(fp+5)
|
||
|
<< " [4]:" << *(fp+4)
|
||
|
<< " [3]:" << *(fp+3)
|
||
|
<< " [2]:" << *(fp+2)
|
||
|
<< " [1]:" << *(fp+1)
|
||
|
<< " [0]:" << *fp;
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access Only, no modifications to elements*/
|
||
|
const float& operator[](int i) const
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 7));
|
||
|
float *fp = (float*)&vec;
|
||
|
return *(fp+i);
|
||
|
}
|
||
|
|
||
|
/* Element Access and Modification*/
|
||
|
float& operator[](int i)
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 7));
|
||
|
float *fp = (float*)&vec;
|
||
|
return *(fp+i);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
/* Miscellaneous */
|
||
|
|
||
|
/* Interleave low order data elements of a and b into destination */
|
||
|
inline F32vec8 unpack_low(const F32vec8 &a, const F32vec8 &b){
|
||
|
return _mm256_unpacklo_ps(a, b); }
|
||
|
|
||
|
/* Interleave high order data elements of a and b into target */
|
||
|
inline F32vec8 unpack_high(const F32vec8 &a, const F32vec8 &b){
|
||
|
return _mm256_unpackhi_ps(a, b); }
|
||
|
|
||
|
/* Move Mask to Integer returns 8 bit mask formed of most significant bits of a */
|
||
|
inline int move_mask(const F32vec8 &a){
|
||
|
return _mm256_movemask_ps(a); }
|
||
|
|
||
|
/* Data Motion Functions */
|
||
|
|
||
|
/* Load Unaligned loadu_ps: Unaligned */
|
||
|
inline void loadu(F32vec8 &a, const float *p){
|
||
|
a = _mm256_loadu_ps(p); }
|
||
|
|
||
|
/* Store Unaligned storeu_ps: Unaligned */
|
||
|
inline void storeu(float *p, const F32vec8 &a){
|
||
|
_mm256_storeu_ps(p, a); }
|
||
|
|
||
|
/* Cacheability Support */
|
||
|
|
||
|
/* Non-Temporal Store */
|
||
|
inline void store_nta(float *p, const F32vec8 &a){
|
||
|
_mm256_stream_ps(p, a); }
|
||
|
|
||
|
/* Conditional moves */
|
||
|
|
||
|
/* Masked load */
|
||
|
inline void maskload(F32vec8 &a, const float *p, const F32vec8 &m){
|
||
|
a = _mm256_maskload_ps(p, _mm256_castps_si256(m)); }
|
||
|
|
||
|
inline void maskload(F32vec4 &a, const float *p, const F32vec4 &m){
|
||
|
a = _mm_maskload_ps(p, _mm_castps_si128(m)); }
|
||
|
|
||
|
/* Masked store */
|
||
|
inline void maskstore(float *p, const F32vec8 &a, const F32vec8 &m){
|
||
|
_mm256_maskstore_ps(p, _mm256_castps_si256(m), a); }
|
||
|
|
||
|
inline void maskstore(float *p, const F32vec4 &a, const F32vec4 &m){
|
||
|
_mm_maskstore_ps(p, _mm_castps_si128(m), a); }
|
||
|
|
||
|
/* Conditional Selects */
|
||
|
|
||
|
inline F32vec8 select_eq(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_EQ_OQ)); }
|
||
|
|
||
|
inline F32vec8 select_lt(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_LT_OS)); }
|
||
|
|
||
|
inline F32vec8 select_le(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_LE_OS)); }
|
||
|
|
||
|
inline F32vec8 select_gt(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_GT_OS)); }
|
||
|
|
||
|
inline F32vec8 select_ge(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_GE_OS)); }
|
||
|
|
||
|
inline F32vec8 select_neq(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_NEQ_UQ)); }
|
||
|
|
||
|
inline F32vec8 select_nlt(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_NLT_US)); }
|
||
|
|
||
|
inline F32vec8 select_nle(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_NLE_US)); }
|
||
|
|
||
|
inline F32vec8 select_ngt(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_NGT_US)); }
|
||
|
|
||
|
inline F32vec8 select_nge(const F32vec8 &a, const F32vec8 &b, const F32vec8 &c, const F32vec8 &d){
|
||
|
return _mm256_blendv_ps(d, c, _mm256_cmp_ps(a, b, _CMP_NGE_US)); }
|
||
|
|
||
|
/*
|
||
|
* class F64vec4
|
||
|
*
|
||
|
* Represents 256-bit vector composed of 4 double precision floating point elements.
|
||
|
*/
|
||
|
class F64vec4
|
||
|
{
|
||
|
protected:
|
||
|
__m256d vec;
|
||
|
|
||
|
public:
|
||
|
|
||
|
/* Constructors: __m256d, 4 doubles */
|
||
|
F64vec4() {}
|
||
|
|
||
|
/* initialize 4 DP FP with __m256d data type */
|
||
|
F64vec4(__m256d m) { vec = m; }
|
||
|
|
||
|
/* initialize 4 DP FPs with 4 doubles */
|
||
|
F64vec4(double d3, double d2, double d1, double d0)
|
||
|
{
|
||
|
vec = _mm256_set_pd(d3,d2,d1,d0);
|
||
|
}
|
||
|
|
||
|
/* Explicitly initialize each of 4 DP FPs with same double */
|
||
|
EXPLICIT F64vec4(double d) { vec = _mm256_set1_pd(d); }
|
||
|
|
||
|
/* Conversion functions */
|
||
|
operator __m256d() const { return vec; }
|
||
|
|
||
|
/* Logical Operators */
|
||
|
friend F64vec4 operator &(const F64vec4 &a, const F64vec4 &b) { return _mm256_and_pd(a,b); }
|
||
|
friend F64vec4 operator |(const F64vec4 &a, const F64vec4 &b) { return _mm256_or_pd(a,b); }
|
||
|
friend F64vec4 operator ^(const F64vec4 &a, const F64vec4 &b) { return _mm256_xor_pd(a,b); }
|
||
|
|
||
|
/* Arithmetic Operators */
|
||
|
friend F64vec4 operator +(const F64vec4 &a, const F64vec4 &b) { return _mm256_add_pd(a,b); }
|
||
|
friend F64vec4 operator -(const F64vec4 &a, const F64vec4 &b) { return _mm256_sub_pd(a,b); }
|
||
|
friend F64vec4 operator *(const F64vec4 &a, const F64vec4 &b) { return _mm256_mul_pd(a,b); }
|
||
|
friend F64vec4 operator /(const F64vec4 &a, const F64vec4 &b) { return _mm256_div_pd(a,b); }
|
||
|
|
||
|
F64vec4& operator +=(const F64vec4 &a) { return *this = _mm256_add_pd(vec,a); }
|
||
|
F64vec4& operator -=(const F64vec4 &a) { return *this = _mm256_sub_pd(vec,a); }
|
||
|
F64vec4& operator *=(const F64vec4 &a) { return *this = _mm256_mul_pd(vec,a); }
|
||
|
F64vec4& operator /=(const F64vec4 &a) { return *this = _mm256_div_pd(vec,a); }
|
||
|
F64vec4& operator &=(const F64vec4 &a) { return *this = _mm256_and_pd(vec,a); }
|
||
|
F64vec4& operator |=(const F64vec4 &a) { return *this = _mm256_or_pd(vec,a); }
|
||
|
F64vec4& operator ^=(const F64vec4 &a) { return *this = _mm256_xor_pd(vec,a); }
|
||
|
|
||
|
/* Horizontal Add */
|
||
|
friend double add_horizontal(const F64vec4 &a)
|
||
|
{
|
||
|
F64vec4 temp = _mm256_add_pd(a, _mm256_permute_pd(a,0x05));
|
||
|
return _mm_cvtsd_f64(_mm_add_sd(_mm256_castpd256_pd128(temp), _mm256_extractf128_pd(temp,1)));
|
||
|
}
|
||
|
|
||
|
/* And Not */
|
||
|
friend F64vec4 andnot(const F64vec4 &a, const F64vec4 &b) { return _mm256_andnot_pd(a,b); }
|
||
|
|
||
|
/* Square Root */
|
||
|
friend F64vec4 sqrt(const F64vec4 &a) { return _mm256_sqrt_pd(a); }
|
||
|
|
||
|
/* Compares: Mask is returned */
|
||
|
friend F64vec4 cmp_eq(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_EQ_OQ); }
|
||
|
friend F64vec4 cmp_lt(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_LT_OS); }
|
||
|
friend F64vec4 cmp_le(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_LE_OS); }
|
||
|
friend F64vec4 cmp_gt(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_GT_OS); }
|
||
|
friend F64vec4 cmp_ge(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_GE_OS); }
|
||
|
friend F64vec4 cmp_neq(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_NEQ_UQ); }
|
||
|
friend F64vec4 cmp_nlt(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_NLT_US); }
|
||
|
friend F64vec4 cmp_nle(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_NLE_US); }
|
||
|
friend F64vec4 cmp_ngt(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_NGT_US); }
|
||
|
friend F64vec4 cmp_nge(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_cmp_pd(a, b, _CMP_NGE_US); }
|
||
|
|
||
|
/* Min and Max */
|
||
|
friend F64vec4 simd_min(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_min_pd(a,b); }
|
||
|
friend F64vec4 simd_max(const F64vec4 &a, const F64vec4 &b)
|
||
|
{ return _mm256_max_pd(a,b); }
|
||
|
|
||
|
/* Absolute value */
|
||
|
friend F64vec4 abs(const F64vec4 &a)
|
||
|
{
|
||
|
static const union
|
||
|
{
|
||
|
int i[8];
|
||
|
__m256d m;
|
||
|
} __f64vec4_abs_mask = { 0xffffffff, 0x7fffffff, 0xffffffff, 0x7fffffff,
|
||
|
0xffffffff, 0x7fffffff, 0xffffffff, 0x7fffffff};
|
||
|
return _mm256_and_pd(a, __f64vec4_abs_mask.m);
|
||
|
}
|
||
|
|
||
|
/* Debug Features */
|
||
|
#if defined (_ENABLE_VEC_DEBUG)
|
||
|
/* Output */
|
||
|
friend DVEC_STD ostream & operator<<(DVEC_STD ostream &os, const F64vec4 &a)
|
||
|
{
|
||
|
/* To use: cout << "Elements of F64vec4 fvec are: " << fvec; */
|
||
|
double *dp = (double*) &a;
|
||
|
os << "[3]:" << *(dp+3)
|
||
|
<< " [2]:" << *(dp+2)
|
||
|
<< " [3]:" << *(dp+1)
|
||
|
<< " [0]:" << *dp;
|
||
|
return os;
|
||
|
}
|
||
|
#endif /* defined (_ENABLE_VEC_DEBUG) */
|
||
|
|
||
|
/* Element Access Only, no modifications to elements */
|
||
|
const double& operator[](int i) const
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 3));
|
||
|
double *dp = (double*)&vec;
|
||
|
return *(dp+i);
|
||
|
}
|
||
|
/* Element Access and Modification*/
|
||
|
double& operator[](int i)
|
||
|
{
|
||
|
/* Assert enabled only during debug /DDEBUG */
|
||
|
_VEC_ASSERT((0 <= i) && (i <= 3));
|
||
|
double *dp = (double*)&vec;
|
||
|
return *(dp+i);
|
||
|
}
|
||
|
};
|
||
|
|
||
|
/* Miscellaneous */
|
||
|
|
||
|
/* Interleave low order data elements of a and b into destination */
|
||
|
inline F64vec4 unpack_low(const F64vec4 &a, const F64vec4 &b){
|
||
|
return _mm256_unpacklo_pd(a, b); }
|
||
|
|
||
|
/* Interleave high order data elements of a and b into target */
|
||
|
inline F64vec4 unpack_high(const F64vec4 &a, const F64vec4 &b){
|
||
|
return _mm256_unpackhi_pd(a, b); }
|
||
|
|
||
|
/* Move Mask to Integer returns 4 bit mask formed of most significant bits of a */
|
||
|
inline int move_mask(const F64vec4 &a){
|
||
|
return _mm256_movemask_pd(a); }
|
||
|
|
||
|
/* Data Motion Functions */
|
||
|
|
||
|
/* Load Unaligned loadu_pd: Unaligned */
|
||
|
inline void loadu(F64vec4 &a, double *p){
|
||
|
a = _mm256_loadu_pd(p); }
|
||
|
|
||
|
/* Store Unaligned storeu_pd: Unaligned */
|
||
|
inline void storeu(double *p, const F64vec4 &a){
|
||
|
_mm256_storeu_pd(p, a); }
|
||
|
|
||
|
/* Cacheability Support */
|
||
|
|
||
|
/* Non-Temporal Store */
|
||
|
inline void store_nta(double *p, const F64vec4 &a){
|
||
|
_mm256_stream_pd(p, a); }
|
||
|
|
||
|
/* Conditional moves */
|
||
|
|
||
|
/* Masked load */
|
||
|
inline void maskload(F64vec4 &a, const double *p, const F64vec4 &m){
|
||
|
a = _mm256_maskload_pd(p, _mm256_castpd_si256(m)); }
|
||
|
|
||
|
inline void maskload(F64vec2 &a, const double *p, const F64vec2 &m){
|
||
|
a = _mm_maskload_pd(p, _mm_castpd_si128(m)); }
|
||
|
|
||
|
/* Masked store */
|
||
|
inline void maskstore(double *p, const F64vec4 &a, const F64vec4 &m){
|
||
|
_mm256_maskstore_pd(p, _mm256_castpd_si256(m), a); }
|
||
|
|
||
|
inline void maskstore(double *p, const F64vec2 &a, const F64vec2 &m){
|
||
|
_mm_maskstore_pd(p, _mm_castpd_si128(m), a); }
|
||
|
|
||
|
/* Conditional Selects */
|
||
|
|
||
|
inline F64vec4 select_eq(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_EQ_OQ)); }
|
||
|
|
||
|
inline F64vec4 select_lt(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_LT_OS)); }
|
||
|
|
||
|
inline F64vec4 select_le(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_LE_OS)); }
|
||
|
|
||
|
inline F64vec4 select_gt(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_GT_OS)); }
|
||
|
|
||
|
inline F64vec4 select_ge(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_GE_OS)); }
|
||
|
|
||
|
inline F64vec4 select_neq(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_NEQ_UQ)); }
|
||
|
|
||
|
inline F64vec4 select_nlt(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_NLT_US)); }
|
||
|
|
||
|
inline F64vec4 select_nle(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_NLE_US)); }
|
||
|
|
||
|
inline F64vec4 select_ngt(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_NGT_US)); }
|
||
|
|
||
|
inline F64vec4 select_nge(const F64vec4 &a, const F64vec4 &b, const F64vec4 &c, const F64vec4 &d){
|
||
|
return _mm256_blendv_pd(d, c, _mm256_cmp_pd(a, b, _CMP_NGE_US)); }
|
||
|
|
||
|
/* Conversion Functions */
|
||
|
|
||
|
/* Convert the 4 SP FP values of a to 4 DP FP values */
|
||
|
inline F64vec4 F32vec4ToF64vec4(const F32vec4 &a){
|
||
|
return _mm256_cvtps_pd(a); }
|
||
|
|
||
|
/* Convert the 4 DP FP values of a to 4 SP FP values */
|
||
|
inline F32vec4 F64vec4ToF32vec8(const F64vec4 &a){
|
||
|
return _mm256_cvtpd_ps(a); }
|
||
|
|
||
|
#undef DVEC_DEFINE_OUTPUT_OPERATORS
|
||
|
#undef DVEC_STD
|
||
|
|
||
|
#ifdef _MSC_VER
|
||
|
#pragma pack(pop)
|
||
|
#endif /* _MSC_VER */
|
||
|
|
||
|
#endif /* defined (_M_CEE_PURE) */
|
||
|
|
||
|
#endif /* RC_INVOKED */
|
||
|
#endif /* _DVEC_H_INCLUDED */
|