include/crt/dvec.h

   1 /**
   2  * This file has no copyright assigned and is placed in the Public Domain.
   3  * This file is part of the w64 mingw-runtime package.
   4  * No warranty is given; refer to the file DISCLAIMER within this package.
   5  */
   6 #ifndef _DVEC_H_INCLUDED
   7 #define _DVEC_H_INCLUDED
   8 #ifndef RC_INVOKED
   9
  10 #if !defined __cplusplus
  11 #error This file is only supported in C++ compilations!
  12 #endif
  13
  14 #include <emmintrin.h>
  15 #include <assert.h>
  16 #include <fvec.h>
  17 #include <crtdefs.h>
  18
  19 #pragma pack(push,_CRT_PACKING)
  20
  21 #if defined(_ENABLE_VEC_DEBUG)
  22 #include <iostream>
  23 #endif
  24
  25 #pragma pack(push,16)
  26
  27 #define EXPLICIT explicit
  28
  29 class I8vec16;
  30 class Is8vec16;
  31 class Iu8vec16;
  32 class I16vec8;
  33 class Is16vec8;
  34 class Iu16vec8;
  35 class I32vec4;
  36 class Is32vec4;
  37 class Iu32vec4;
  38 class I64vec2;
  39 class I128vec1;
  40
  41 #define _MM_16UB(element,vector) (*((unsigned char*)&##vector + ##element))
  42 #define _MM_16B(element,vector) (*((signed char*)&##vector + ##element))
  43
  44 #define _MM_8UW(element,vector) (*((unsigned short*)&##vector + ##element))
  45 #define _MM_8W(element,vector) (*((short*)&##vector + ##element))
  46
  47 #define _MM_4UDW(element,vector) (*((unsigned int*)&##vector + ##element))
  48 #define _MM_4DW(element,vector) (*((int*)&##vector + ##element))
  49
  50 #define _MM_2QW(element,vector) (*((__int64*)&##vector + ##element))
  51
  52 inline const __m128i get_mask128()
  53 {
  54   static const __m128i mask128 = _mm_set1_epi64(M64(0xffffffffffffffffi64));
  55   return mask128;
  56 }
  57
  58 class M128
  59 {
  60 protected:
  61   __m128i vec;
  62
  63 public:
  64   M128() { }
  65   M128(__m128i mm) { vec = mm; }
  66
  67   operator __m128i() const { return vec; }
  68
  69   M128& operator&=(const M128 &a) { return *this = (M128) _mm_and_si128(vec,a); }
  70   M128& operator|=(const M128 &a) { return *this = (M128) _mm_or_si128(vec,a); }
  71   M128& operator^=(const M128 &a) { return *this = (M128) _mm_xor_si128(vec,a); }
  72
  73 };
  74
  75 inline M128 operator&(const M128 &a,const M128 &b) { return _mm_and_si128(a,b); }
  76 inline M128 operator|(const M128 &a,const M128 &b) { return _mm_or_si128(a,b); }
  77 inline M128 operator^(const M128 &a,const M128 &b) { return _mm_xor_si128(a,b); }
  78 inline M128 andnot(const M128 &a,const M128 &b) { return _mm_andnot_si128(a,b); }
  79
  80 class I128vec1 : public M128
  81 {
  82 public:
  83   I128vec1() { }
  84   I128vec1(__m128i mm) : M128(mm) { }
  85
  86   I128vec1& operator= (const M128 &a) { return *this = (I128vec1) a; }
  87   I128vec1& operator&=(const M128 &a) { return *this = (I128vec1) _mm_and_si128(vec,a); }
  88   I128vec1& operator|=(const M128 &a) { return *this = (I128vec1) _mm_or_si128(vec,a); }
  89   I128vec1& operator^=(const M128 &a) { return *this = (I128vec1) _mm_xor_si128(vec,a); }
  90
  91 };
  92
  93 class I64vec2 : public M128
  94 {
  95 public:
  96   I64vec2() { }
  97   I64vec2(__m128i mm) : M128(mm) { }
  98
  99   I64vec2(__m64 q1,__m64 q0)
 100   {
 101     _MM_2QW(0,vec) = *(__int64*)&q0;
 102     _MM_2QW(1,vec) = *(__int64*)&q1;
 103   }
 104
 105   I64vec2& operator= (const M128 &a) { return *this = (I64vec2) a; }
 106
 107   I64vec2& operator&=(const M128 &a) { return *this = (I64vec2) _mm_and_si128(vec,a); }
 108   I64vec2& operator|=(const M128 &a) { return *this = (I64vec2) _mm_or_si128(vec,a); }
 109   I64vec2& operator^=(const M128 &a) { return *this = (I64vec2) _mm_xor_si128(vec,a); }
 110
 111   I64vec2& operator +=(const I64vec2 &a) { return *this = (I64vec2) _mm_add_epi64(vec,a); }
 112   I64vec2& operator -=(const I64vec2 &a) { return *this = (I64vec2) _mm_sub_epi64(vec,a); }
 113
 114   I64vec2 operator<<(const I64vec2 &a) { return _mm_sll_epi64(vec,a); }
 115   I64vec2 operator<<(int count) { return _mm_slli_epi64(vec,count); }
 116   I64vec2& operator<<=(const I64vec2 &a) { return *this = (I64vec2) _mm_sll_epi64(vec,a); }
 117   I64vec2& operator<<=(int count) { return *this = (I64vec2) _mm_slli_epi64(vec,count); }
 118   I64vec2 operator>>(const I64vec2 &a) { return _mm_srl_epi64(vec,a); }
 119   I64vec2 operator>>(int count) { return _mm_srli_epi64(vec,count); }
 120   I64vec2& operator>>=(const I64vec2 &a) { return *this = (I64vec2) _mm_srl_epi64(vec,a); }
 121   I64vec2& operator>>=(int count) { return *this = (I64vec2) _mm_srli_epi64(vec,count); }
 122
 123   const __int64& operator[](int i)const
 124   {
 125     assert(static_cast<unsigned int>(i) < 2);
 126     return _MM_2QW(i,vec);
 127   }
 128
 129   __int64& operator[](int i)
 130   {
 131     assert(static_cast<unsigned int>(i) < 2);
 132     return _MM_2QW(i,vec);
 133   }
 134
 135 };
 136
 137 inline I64vec2 unpack_low(const I64vec2 &a,const I64vec2 &b) {return _mm_unpacklo_epi64(a,b); }
 138 inline I64vec2 unpack_high(const I64vec2 &a,const I64vec2 &b) {return _mm_unpackhi_epi64(a,b); }
 139
 140 class I32vec4 : public M128
 141 {
 142 public:
 143   I32vec4() { }
 144   I32vec4(__m128i mm) : M128(mm) { }
 145
 146   I32vec4& operator= (const M128 &a) { return *this = (I32vec4) a; }
 147
 148   I32vec4& operator&=(const M128 &a) { return *this = (I32vec4) _mm_and_si128(vec,a); }
 149   I32vec4& operator|=(const M128 &a) { return *this = (I32vec4) _mm_or_si128(vec,a); }
 150   I32vec4& operator^=(const M128 &a) { return *this = (I32vec4) _mm_xor_si128(vec,a); }
 151
 152   I32vec4& operator +=(const I32vec4 &a) { return *this = (I32vec4)_mm_add_epi32(vec,a); }
 153   I32vec4& operator -=(const I32vec4 &a) { return *this = (I32vec4)_mm_sub_epi32(vec,a); }
 154
 155   I32vec4 operator<<(const I32vec4 &a) { return _mm_sll_epi32(vec,a); }
 156   I32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
 157   I32vec4& operator<<=(const I32vec4 &a) { return *this = (I32vec4)_mm_sll_epi32(vec,a); }
 158   I32vec4& operator<<=(int count) { return *this = (I32vec4)_mm_slli_epi32(vec,count); }
 159
 160 };
 161
 162 inline I32vec4 cmpeq(const I32vec4 &a,const I32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
 163 inline I32vec4 cmpneq(const I32vec4 &a,const I32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b),get_mask128()); }
 164
 165 inline I32vec4 unpack_low(const I32vec4 &a,const I32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
 166 inline I32vec4 unpack_high(const I32vec4 &a,const I32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
 167
 168 class Is32vec4 : public I32vec4
 169 {
 170 public:
 171   Is32vec4() { }
 172   Is32vec4(__m128i mm) : I32vec4(mm) { }
 173   Is32vec4(int i3,int i2,int i1,int i0)
 174   {
 175     _MM_4DW(0,vec) = i0;
 176     _MM_4DW(1,vec) = i1;
 177     _MM_4DW(2,vec) = i2;
 178     _MM_4DW(3,vec) = i3;
 179   }
 180
 181   Is32vec4& operator= (const M128 &a) { return *this = (Is32vec4) a; }
 182
 183   Is32vec4& operator&=(const M128 &a) { return *this = (Is32vec4) _mm_and_si128(vec,a); }
 184   Is32vec4& operator|=(const M128 &a) { return *this = (Is32vec4) _mm_or_si128(vec,a); }
 185   Is32vec4& operator^=(const M128 &a) { return *this = (Is32vec4) _mm_xor_si128(vec,a); }
 186
 187   Is32vec4& operator +=(const I32vec4 &a) { return *this = (Is32vec4)_mm_add_epi32(vec,a); }
 188   Is32vec4& operator -=(const I32vec4 &a) { return *this = (Is32vec4)_mm_sub_epi32(vec,a); }
 189
 190   Is32vec4 operator<<(const M128 &a) { return _mm_sll_epi32(vec,a); }
 191   Is32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
 192   Is32vec4& operator<<=(const M128 &a) { return *this = (Is32vec4)_mm_sll_epi32(vec,a); }
 193   Is32vec4& operator<<=(int count) { return *this = (Is32vec4)_mm_slli_epi32(vec,count); }
 194
 195   Is32vec4 operator>>(const M128 &a) { return _mm_sra_epi32(vec,a); }
 196   Is32vec4 operator>>(int count) { return _mm_srai_epi32(vec,count); }
 197   Is32vec4& operator>>=(const M128 &a) { return *this = (Is32vec4) _mm_sra_epi32(vec,a); }
 198   Is32vec4& operator>>=(int count) { return *this = (Is32vec4) _mm_srai_epi32(vec,count); }
 199
 200 #if defined(_ENABLE_VEC_DEBUG)
 201
 202   friend std::ostream& operator<< (std::ostream &os,const Is32vec4 &a)
 203   {
 204     os << "[3]:" << _MM_4DW(3,a)
 205       << " [2]:" << _MM_4DW(2,a)
 206       << " [1]:" << _MM_4DW(1,a)
 207       << " [0]:" << _MM_4DW(0,a);
 208     return os;
 209   }
 210 #endif
 211
 212   const int& operator[](int i)const
 213   {
 214     assert(static_cast<unsigned int>(i) < 4);
 215     return _MM_4DW(i,vec);
 216   }
 217
 218   int& operator[](int i)
 219   {
 220     assert(static_cast<unsigned int>(i) < 4);
 221     return _MM_4DW(i,vec);
 222   }
 223 };
 224
 225 inline Is32vec4 cmpeq(const Is32vec4 &a,const Is32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
 226 inline Is32vec4 cmpneq(const Is32vec4 &a,const Is32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b),get_mask128()); }
 227 inline Is32vec4 cmpgt(const Is32vec4 &a,const Is32vec4 &b) { return _mm_cmpgt_epi32(a,b); }
 228 inline Is32vec4 cmplt(const Is32vec4 &a,const Is32vec4 &b) { return _mm_cmpgt_epi32(b,a); }
 229
 230 inline Is32vec4 unpack_low(const Is32vec4 &a,const Is32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
 231 inline Is32vec4 unpack_high(const Is32vec4 &a,const Is32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
 232
 233 class Iu32vec4 : public I32vec4
 234 {
 235 public:
 236   Iu32vec4() { }
 237   Iu32vec4(__m128i mm) : I32vec4(mm) { }
 238   Iu32vec4(unsigned int ui3,unsigned int ui2,unsigned int ui1,unsigned int ui0)
 239   {
 240     _MM_4UDW(0,vec) = ui0;
 241     _MM_4UDW(1,vec) = ui1;
 242     _MM_4UDW(2,vec) = ui2;
 243     _MM_4UDW(3,vec) = ui3;
 244   }
 245
 246   Iu32vec4& operator= (const M128 &a) { return *this = (Iu32vec4) a; }
 247
 248   Iu32vec4& operator&=(const M128 &a) { return *this = (Iu32vec4) _mm_and_si128(vec,a); }
 249   Iu32vec4& operator|=(const M128 &a) { return *this = (Iu32vec4) _mm_or_si128(vec,a); }
 250   Iu32vec4& operator^=(const M128 &a) { return *this = (Iu32vec4) _mm_xor_si128(vec,a); }
 251
 252   Iu32vec4& operator +=(const I32vec4 &a) { return *this = (Iu32vec4)_mm_add_epi32(vec,a); }
 253   Iu32vec4& operator -=(const I32vec4 &a) { return *this = (Iu32vec4)_mm_sub_epi32(vec,a); }
 254
 255   Iu32vec4 operator<<(const M128 &a) { return _mm_sll_epi32(vec,a); }
 256   Iu32vec4 operator<<(int count) { return _mm_slli_epi32(vec,count); }
 257   Iu32vec4& operator<<=(const M128 &a) { return *this = (Iu32vec4)_mm_sll_epi32(vec,a); }
 258   Iu32vec4& operator<<=(int count) { return *this = (Iu32vec4)_mm_slli_epi32(vec,count); }
 259   Iu32vec4 operator>>(const M128 &a) { return _mm_srl_epi32(vec,a); }
 260   Iu32vec4 operator>>(int count) { return _mm_srli_epi32(vec,count); }
 261   Iu32vec4& operator>>=(const M128 &a) { return *this = (Iu32vec4) _mm_srl_epi32(vec,a); }
 262   Iu32vec4& operator>>=(int count) { return *this = (Iu32vec4) _mm_srli_epi32(vec,count); }
 263
 264 #if defined(_ENABLE_VEC_DEBUG)
 265
 266   friend std::ostream& operator<< (std::ostream &os,const Iu32vec4 &a)
 267   {
 268     os << "[3]:" << _MM_4UDW(3,a)
 269       << " [2]:" << _MM_4UDW(2,a)
 270       << " [1]:" << _MM_4UDW(1,a)
 271       << " [0]:" << _MM_4UDW(0,a);
 272     return os;
 273   }
 274 #endif
 275
 276   const unsigned int& operator[](int i)const
 277   {
 278     assert(static_cast<unsigned int>(i) < 4);
 279     return _MM_4UDW(i,vec);
 280   }
 281
 282   unsigned int& operator[](int i)
 283   {
 284     assert(static_cast<unsigned int>(i) < 4);
 285     return _MM_4UDW(i,vec);
 286   }
 287 };
 288
 289 inline I64vec2 operator*(const Iu32vec4 &a,const Iu32vec4 &b) { return _mm_mul_epu32(a,b); }
 290 inline Iu32vec4 cmpeq(const Iu32vec4 &a,const Iu32vec4 &b) { return _mm_cmpeq_epi32(a,b); }
 291 inline Iu32vec4 cmpneq(const Iu32vec4 &a,const Iu32vec4 &b) { return _mm_andnot_si128(_mm_cmpeq_epi32(a,b),get_mask128()); }
 292
 293 inline Iu32vec4 unpack_low(const Iu32vec4 &a,const Iu32vec4 &b) { return _mm_unpacklo_epi32(a,b); }
 294 inline Iu32vec4 unpack_high(const Iu32vec4 &a,const Iu32vec4 &b) { return _mm_unpackhi_epi32(a,b); }
 295
 296 class I16vec8 : public M128
 297 {
 298 public:
 299   I16vec8() { }
 300   I16vec8(__m128i mm) : M128(mm) { }
 301
 302   I16vec8& operator= (const M128 &a) { return *this = (I16vec8) a; }
 303
 304   I16vec8& operator&=(const M128 &a) { return *this = (I16vec8) _mm_and_si128(vec,a); }
 305   I16vec8& operator|=(const M128 &a) { return *this = (I16vec8) _mm_or_si128(vec,a); }
 306   I16vec8& operator^=(const M128 &a) { return *this = (I16vec8) _mm_xor_si128(vec,a); }
 307
 308   I16vec8& operator +=(const I16vec8 &a) { return *this = (I16vec8) _mm_add_epi16(vec,a); }
 309   I16vec8& operator -=(const I16vec8 &a) { return *this = (I16vec8) _mm_sub_epi16(vec,a); }
 310   I16vec8& operator *=(const I16vec8 &a) { return *this = (I16vec8) _mm_mullo_epi16(vec,a); }
 311
 312   I16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
 313   I16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
 314   I16vec8& operator<<=(const M128 &a) { return *this = (I16vec8)_mm_sll_epi16(vec,a); }
 315   I16vec8& operator<<=(int count) { return *this = (I16vec8)_mm_slli_epi16(vec,count); }
 316
 317 };
 318
 319 inline I16vec8 operator*(const I16vec8 &a,const I16vec8 &b) { return _mm_mullo_epi16(a,b); }
 320
 321 inline I16vec8 cmpeq(const I16vec8 &a,const I16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
 322 inline I16vec8 cmpneq(const I16vec8 &a,const I16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b),get_mask128()); }
 323
 324 inline I16vec8 unpack_low(const I16vec8 &a,const I16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
 325 inline I16vec8 unpack_high(const I16vec8 &a,const I16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
 326
 327 class Is16vec8 : public I16vec8
 328 {
 329 public:
 330   Is16vec8() { }
 331   Is16vec8(__m128i mm) : I16vec8(mm) { }
 332   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)
 333   {
 334     _MM_8W(0,vec) = s0;
 335     _MM_8W(1,vec) = s1;
 336     _MM_8W(2,vec) = s2;
 337     _MM_8W(3,vec) = s3;
 338     _MM_8W(4,vec) = s4;
 339     _MM_8W(5,vec) = s5;
 340     _MM_8W(6,vec) = s6;
 341     _MM_8W(7,vec) = s7;
 342   }
 343
 344   Is16vec8& operator= (const M128 &a) { return *this = (Is16vec8) a; }
 345
 346   Is16vec8& operator&=(const M128 &a) { return *this = (Is16vec8) _mm_and_si128(vec,a); }
 347   Is16vec8& operator|=(const M128 &a) { return *this = (Is16vec8) _mm_or_si128(vec,a); }
 348   Is16vec8& operator^=(const M128 &a) { return *this = (Is16vec8) _mm_xor_si128(vec,a); }
 349
 350   Is16vec8& operator +=(const I16vec8 &a) { return *this = (Is16vec8) _mm_add_epi16(vec,a); }
 351   Is16vec8& operator -=(const I16vec8 &a) { return *this = (Is16vec8) _mm_sub_epi16(vec,a); }
 352   Is16vec8& operator *=(const I16vec8 &a) { return *this = (Is16vec8) _mm_mullo_epi16(vec,a); }
 353
 354   Is16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
 355   Is16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
 356   Is16vec8& operator<<=(const M128 &a) { return *this = (Is16vec8)_mm_sll_epi16(vec,a); }
 357   Is16vec8& operator<<=(int count) { return *this = (Is16vec8)_mm_slli_epi16(vec,count); }
 358
 359   Is16vec8 operator>>(const M128 &a) { return _mm_sra_epi16(vec,a); }
 360   Is16vec8 operator>>(int count) { return _mm_srai_epi16(vec,count); }
 361   Is16vec8& operator>>=(const M128 &a) { return *this = (Is16vec8)_mm_sra_epi16(vec,a); }
 362   Is16vec8& operator>>=(int count) { return *this = (Is16vec8)_mm_srai_epi16(vec,count); }
 363
 364 #if defined(_ENABLE_VEC_DEBUG)
 365
 366   friend std::ostream& operator<< (std::ostream &os,const Is16vec8 &a)
 367   {
 368     os << "[7]:" << _MM_8W(7,a)
 369       << " [6]:" << _MM_8W(6,a)
 370       << " [5]:" << _MM_8W(5,a)
 371       << " [4]:" << _MM_8W(4,a)
 372       << " [3]:" << _MM_8W(3,a)
 373       << " [2]:" << _MM_8W(2,a)
 374       << " [1]:" << _MM_8W(1,a)
 375       << " [0]:" << _MM_8W(0,a);
 376     return os;
 377   }
 378 #endif
 379
 380   const signed short& operator[](int i)const
 381   {
 382     assert(static_cast<unsigned int>(i) < 8);
 383     return _MM_8W(i,vec);
 384   }
 385
 386   signed short& operator[](int i)
 387   {
 388     assert(static_cast<unsigned int>(i) < 8);
 389     return _MM_8W(i,vec);
 390   }
 391 };
 392
 393 inline Is16vec8 operator*(const Is16vec8 &a,const Is16vec8 &b) { return _mm_mullo_epi16(a,b); }
 394
 395 inline Is16vec8 cmpeq(const Is16vec8 &a,const Is16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
 396 inline Is16vec8 cmpneq(const Is16vec8 &a,const Is16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b),get_mask128()); }
 397 inline Is16vec8 cmpgt(const Is16vec8 &a,const Is16vec8 &b) { return _mm_cmpgt_epi16(a,b); }
 398 inline Is16vec8 cmplt(const Is16vec8 &a,const Is16vec8 &b) { return _mm_cmpgt_epi16(b,a); }
 399
 400 inline Is16vec8 unpack_low(const Is16vec8 &a,const Is16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
 401 inline Is16vec8 unpack_high(const Is16vec8 &a,const Is16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
 402
 403 inline Is16vec8 mul_high(const Is16vec8 &a,const Is16vec8 &b) { return _mm_mulhi_epi16(a,b); }
 404 inline Is32vec4 mul_add(const Is16vec8 &a,const Is16vec8 &b) { return _mm_madd_epi16(a,b);}
 405
 406 inline Is16vec8 sat_add(const Is16vec8 &a,const Is16vec8 &b) { return _mm_adds_epi16(a,b); }
 407 inline Is16vec8 sat_sub(const Is16vec8 &a,const Is16vec8 &b) { return _mm_subs_epi16(a,b); }
 408
 409 inline Is16vec8 simd_max(const Is16vec8 &a,const Is16vec8 &b) { return _mm_max_epi16(a,b); }
 410 inline Is16vec8 simd_min(const Is16vec8 &a,const Is16vec8 &b) { return _mm_min_epi16(a,b); }
 411
 412 class Iu16vec8 : public I16vec8
 413 {
 414 public:
 415   Iu16vec8() { }
 416   Iu16vec8(__m128i mm) : I16vec8(mm) { }
 417   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)
 418   {
 419     _MM_8UW(0,vec) = s0;
 420     _MM_8UW(1,vec) = s1;
 421     _MM_8UW(2,vec) = s2;
 422     _MM_8UW(3,vec) = s3;
 423     _MM_8UW(4,vec) = s4;
 424     _MM_8UW(5,vec) = s5;
 425     _MM_8UW(6,vec) = s6;
 426     _MM_8UW(7,vec) = s7;
 427   }
 428
 429   Iu16vec8& operator= (const M128 &a) { return *this = (Iu16vec8) a; }
 430
 431   Iu16vec8& operator&=(const M128 &a) { return *this = (Iu16vec8) _mm_and_si128(vec,a); }
 432   Iu16vec8& operator|=(const M128 &a) { return *this = (Iu16vec8) _mm_or_si128(vec,a); }
 433   Iu16vec8& operator^=(const M128 &a) { return *this = (Iu16vec8) _mm_xor_si128(vec,a); }
 434
 435   Iu16vec8& operator +=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_add_epi16(vec,a); }
 436   Iu16vec8& operator -=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_sub_epi16(vec,a); }
 437   Iu16vec8& operator *=(const I16vec8 &a) { return *this = (Iu16vec8) _mm_mullo_epi16(vec,a); }
 438
 439   Iu16vec8 operator<<(const M128 &a) { return _mm_sll_epi16(vec,a); }
 440   Iu16vec8 operator<<(int count) { return _mm_slli_epi16(vec,count); }
 441   Iu16vec8& operator<<=(const M128 &a) { return *this = (Iu16vec8)_mm_sll_epi16(vec,a); }
 442   Iu16vec8& operator<<=(int count) { return *this = (Iu16vec8)_mm_slli_epi16(vec,count); }
 443   Iu16vec8 operator>>(const M128 &a) { return _mm_srl_epi16(vec,a); }
 444   Iu16vec8 operator>>(int count) { return _mm_srli_epi16(vec,count); }
 445   Iu16vec8& operator>>=(const M128 &a) { return *this = (Iu16vec8) _mm_srl_epi16(vec,a); }
 446   Iu16vec8& operator>>=(int count) { return *this = (Iu16vec8) _mm_srli_epi16(vec,count); }
 447
 448 #if defined(_ENABLE_VEC_DEBUG)
 449
 450   friend std::ostream& operator << (std::ostream &os,const Iu16vec8 &a)
 451   {
 452     os << "[7]:" << unsigned short(_MM_8UW(7,a))
 453       << " [6]:" << unsigned short(_MM_8UW(6,a))
 454       << " [5]:" << unsigned short(_MM_8UW(5,a))
 455       << " [4]:" << unsigned short(_MM_8UW(4,a))
 456       << " [3]:" << unsigned short(_MM_8UW(3,a))
 457       << " [2]:" << unsigned short(_MM_8UW(2,a))
 458       << " [1]:" << unsigned short(_MM_8UW(1,a))
 459       << " [0]:" << unsigned short(_MM_8UW(0,a));
 460     return os;
 461   }
 462 #endif
 463
 464   const unsigned short& operator[](int i)const
 465   {
 466     assert(static_cast<unsigned int>(i) < 8);
 467     return _MM_8UW(i,vec);
 468   }
 469
 470   unsigned short& operator[](int i)
 471   {
 472     assert(static_cast<unsigned int>(i) < 8);
 473     return _MM_8UW(i,vec);
 474   }
 475 };
 476
 477 inline Iu16vec8 operator*(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_mullo_epi16(a,b); }
 478
 479 inline Iu16vec8 cmpeq(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_cmpeq_epi16(a,b); }
 480 inline Iu16vec8 cmpneq(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_andnot_si128(_mm_cmpeq_epi16(a,b),get_mask128()); }
 481
 482 inline Iu16vec8 unpack_low(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_unpacklo_epi16(a,b); }
 483 inline Iu16vec8 unpack_high(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_unpackhi_epi16(a,b); }
 484
 485 inline Iu16vec8 sat_add(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_adds_epu16(a,b); }
 486 inline Iu16vec8 sat_sub(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_subs_epu16(a,b); }
 487
 488 inline Iu16vec8 simd_avg(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_avg_epu16(a,b); }
 489 inline I16vec8 mul_high(const Iu16vec8 &a,const Iu16vec8 &b) { return _mm_mulhi_epu16(a,b); }
 490
 491 class I8vec16 : public M128
 492 {
 493 public:
 494   I8vec16() { }
 495   I8vec16(__m128i mm) : M128(mm) { }
 496
 497   I8vec16& operator= (const M128 &a) { return *this = (I8vec16) a; }
 498
 499   I8vec16& operator&=(const M128 &a) { return *this = (I8vec16) _mm_and_si128(vec,a); }
 500   I8vec16& operator|=(const M128 &a) { return *this = (I8vec16) _mm_or_si128(vec,a); }
 501   I8vec16& operator^=(const M128 &a) { return *this = (I8vec16) _mm_xor_si128(vec,a); }
 502
 503   I8vec16& operator +=(const I8vec16 &a) { return *this = (I8vec16) _mm_add_epi8(vec,a); }
 504   I8vec16& operator -=(const I8vec16 &a) { return *this = (I8vec16) _mm_sub_epi8(vec,a); }
 505
 506 };
 507
 508 inline I8vec16 cmpeq(const I8vec16 &a,const I8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
 509 inline I8vec16 cmpneq(const I8vec16 &a,const I8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b),get_mask128()); }
 510
 511 inline I8vec16 unpack_low(const I8vec16 &a,const I8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
 512 inline I8vec16 unpack_high(const I8vec16 &a,const I8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
 513
 514 class Is8vec16 : public I8vec16
 515 {
 516 public:
 517   Is8vec16() { }
 518   Is8vec16(__m128i mm) : I8vec16(mm) { }
 519
 520   Is8vec16& operator= (const M128 &a) { return *this = (Is8vec16) a; }
 521
 522   Is8vec16& operator&=(const M128 &a) { return *this = (Is8vec16) _mm_and_si128(vec,a); }
 523   Is8vec16& operator|=(const M128 &a) { return *this = (Is8vec16) _mm_or_si128(vec,a); }
 524   Is8vec16& operator^=(const M128 &a) { return *this = (Is8vec16) _mm_xor_si128(vec,a); }
 525
 526   Is8vec16& operator +=(const I8vec16 &a) { return *this = (Is8vec16) _mm_add_epi8(vec,a); }
 527   Is8vec16& operator -=(const I8vec16 &a) { return *this = (Is8vec16) _mm_sub_epi8(vec,a); }
 528
 529 #if defined(_ENABLE_VEC_DEBUG)
 530
 531   friend std::ostream& operator << (std::ostream &os,const Is8vec16 &a)
 532   {
 533     os << "[15]:" << short(_MM_16B(15,a))
 534       << " [14]:" << short(_MM_16B(14,a))
 535       << " [13]:" << short(_MM_16B(13,a))
 536       << " [12]:" << short(_MM_16B(12,a))
 537       << " [11]:" << short(_MM_16B(11,a))
 538       << " [10]:" << short(_MM_16B(10,a))
 539       << " [9]:" << short(_MM_16B(9,a))
 540       << " [8]:" << short(_MM_16B(8,a))
 541       << " [7]:" << short(_MM_16B(7,a))
 542       << " [6]:" << short(_MM_16B(6,a))
 543       << " [5]:" << short(_MM_16B(5,a))
 544       << " [4]:" << short(_MM_16B(4,a))
 545       << " [3]:" << short(_MM_16B(3,a))
 546       << " [2]:" << short(_MM_16B(2,a))
 547       << " [1]:" << short(_MM_16B(1,a))
 548       << " [0]:" << short(_MM_16B(0,a));
 549     return os;
 550   }
 551 #endif
 552
 553   const signed char& operator[](int i)const
 554   {
 555     assert(static_cast<unsigned int>(i) < 16);
 556     return _MM_16B(i,vec);
 557   }
 558
 559   signed char& operator[](int i)
 560   {
 561     assert(static_cast<unsigned int>(i) < 16);
 562     return _MM_16B(i,vec);
 563   }
 564
 565 };
 566
 567 inline Is8vec16 cmpeq(const Is8vec16 &a,const Is8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
 568 inline Is8vec16 cmpneq(const Is8vec16 &a,const Is8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b),get_mask128()); }
 569 inline Is8vec16 cmpgt(const Is8vec16 &a,const Is8vec16 &b) { return _mm_cmpgt_epi8(a,b); }
 570 inline Is8vec16 cmplt(const Is8vec16 &a,const Is8vec16 &b) { return _mm_cmplt_epi8(a,b); }
 571
 572 inline Is8vec16 unpack_low(const Is8vec16 &a,const Is8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
 573 inline Is8vec16 unpack_high(const Is8vec16 &a,const Is8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
 574
 575 inline Is8vec16 sat_add(const Is8vec16 &a,const Is8vec16 &b) { return _mm_adds_epi8(a,b); }
 576 inline Is8vec16 sat_sub(const Is8vec16 &a,const Is8vec16 &b) { return _mm_subs_epi8(a,b); }
 577
 578 class Iu8vec16 : public I8vec16
 579 {
 580 public:
 581   Iu8vec16() { }
 582   Iu8vec16(__m128i mm) : I8vec16(mm) { }
 583
 584   Iu8vec16& operator= (const M128 &a) { return *this = (Iu8vec16) a; }
 585
 586   Iu8vec16& operator&=(const M128 &a) { return *this = (Iu8vec16) _mm_and_si128(vec,a); }
 587   Iu8vec16& operator|=(const M128 &a) { return *this = (Iu8vec16) _mm_or_si128(vec,a); }
 588   Iu8vec16& operator^=(const M128 &a) { return *this = (Iu8vec16) _mm_xor_si128(vec,a); }
 589
 590   Iu8vec16& operator +=(const I8vec16 &a) { return *this = (Iu8vec16) _mm_add_epi8(vec,a); }
 591   Iu8vec16& operator -=(const I8vec16 &a) { return *this = (Iu8vec16) _mm_sub_epi8(vec,a); }
 592
 593 #if defined(_ENABLE_VEC_DEBUG)
 594
 595   friend std::ostream& operator << (std::ostream &os,const Iu8vec16 &a)
 596   {
 597     os << "[15]:" << unsigned short(_MM_16UB(15,a))
 598       << " [14]:" << unsigned short(_MM_16UB(14,a))
 599       << " [13]:" << unsigned short(_MM_16UB(13,a))
 600       << " [12]:" << unsigned short(_MM_16UB(12,a))
 601       << " [11]:" << unsigned short(_MM_16UB(11,a))
 602       << " [10]:" << unsigned short(_MM_16UB(10,a))
 603       << " [9]:" << unsigned short(_MM_16UB(9,a))
 604       << " [8]:" << unsigned short(_MM_16UB(8,a))
 605       << " [7]:" << unsigned short(_MM_16UB(7,a))
 606       << " [6]:" << unsigned short(_MM_16UB(6,a))
 607       << " [5]:" << unsigned short(_MM_16UB(5,a))
 608       << " [4]:" << unsigned short(_MM_16UB(4,a))
 609       << " [3]:" << unsigned short(_MM_16UB(3,a))
 610       << " [2]:" << unsigned short(_MM_16UB(2,a))
 611       << " [1]:" << unsigned short(_MM_16UB(1,a))
 612       << " [0]:" << unsigned short(_MM_16UB(0,a));
 613     return os;
 614   }
 615 #endif
 616
 617   const unsigned char& operator[](int i)const
 618   {
 619     assert(static_cast<unsigned int>(i) < 16);
 620     return _MM_16UB(i,vec);
 621   }
 622
 623   unsigned char& operator[](int i)
 624   {
 625     assert(static_cast<unsigned int>(i) < 16);
 626     return _MM_16UB(i,vec);
 627   }
 628
 629 };
 630
 631 inline Iu8vec16 cmpeq(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_cmpeq_epi8(a,b); }
 632 inline Iu8vec16 cmpneq(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_andnot_si128(_mm_cmpeq_epi8(a,b),get_mask128()); }
 633
 634 inline Iu8vec16 unpack_low(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_unpacklo_epi8(a,b); }
 635 inline Iu8vec16 unpack_high(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_unpackhi_epi8(a,b); }
 636
 637 inline Iu8vec16 sat_add(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_adds_epu8(a,b); }
 638 inline Iu8vec16 sat_sub(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_subs_epu8(a,b); }
 639
 640 inline I64vec2 sum_abs(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_sad_epu8(a,b); }
 641
 642 inline Iu8vec16 simd_avg(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_avg_epu8(a,b); }
 643 inline Iu8vec16 simd_max(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_max_epu8(a,b); }
 644 inline Iu8vec16 simd_min(const Iu8vec16 &a,const Iu8vec16 &b) { return _mm_min_epu8(a,b); }
 645
 646 inline Is16vec8 pack_sat(const Is32vec4 &a,const Is32vec4 &b) { return _mm_packs_epi32(a,b); }
 647 inline Is8vec16 pack_sat(const Is16vec8 &a,const Is16vec8 &b) { return _mm_packs_epi16(a,b); }
 648 inline Iu8vec16 packu_sat(const Is16vec8 &a,const Is16vec8 &b) { return _mm_packus_epi16(a,b);}
 649
 650 #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); }
 651
 652 IVEC128_LOGICALS(8,16)
 653 IVEC128_LOGICALS(u8,16)
 654 IVEC128_LOGICALS(s8,16)
 655 IVEC128_LOGICALS(16,8)
 656 IVEC128_LOGICALS(u16,8)
 657 IVEC128_LOGICALS(s16,8)
 658 IVEC128_LOGICALS(32,4)
 659 IVEC128_LOGICALS(u32,4)
 660 IVEC128_LOGICALS(s32,4)
 661 IVEC128_LOGICALS(64,2)
 662 IVEC128_LOGICALS(128,1)
 663 #undef IVEC128_LOGICALS
 664
 665 #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); }
 666
 667 IVEC128_ADD_SUB(8,16,epi8)
 668 IVEC128_ADD_SUB(u8,16,epi8)
 669 IVEC128_ADD_SUB(s8,16,epi8)
 670 IVEC128_ADD_SUB(16,8,epi16)
 671 IVEC128_ADD_SUB(u16,8,epi16)
 672 IVEC128_ADD_SUB(s16,8,epi16)
 673 IVEC128_ADD_SUB(32,4,epi32)
 674 IVEC128_ADD_SUB(u32,4,epi32)
 675 IVEC128_ADD_SUB(s32,4,epi32)
 676 IVEC128_ADD_SUB(64,2,epi64)
 677 #undef IVEC128_ADD_SUB
 678
 679 #define IVEC128_SELECT(vect12,vect34,element,selop,arg1,arg2) 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 & arg1) | I##vect12##vec##element ((_mm_andnot_si128(mask,arg2))))); }
 680 IVEC128_SELECT(8,s8,16,eq,c,d)
 681 IVEC128_SELECT(8,u8,16,eq,c,d)
 682 IVEC128_SELECT(8,8,16,eq,c,d)
 683 IVEC128_SELECT(8,s8,16,neq,c,d)
 684 IVEC128_SELECT(8,u8,16,neq,c,d)
 685 IVEC128_SELECT(8,8,16,neq,c,d)
 686
 687 IVEC128_SELECT(16,s16,8,eq,c,d)
 688 IVEC128_SELECT(16,u16,8,eq,c,d)
 689 IVEC128_SELECT(16,16,8,eq,c,d)
 690 IVEC128_SELECT(16,s16,8,neq,c,d)
 691 IVEC128_SELECT(16,u16,8,neq,c,d)
 692 IVEC128_SELECT(16,16,8,neq,c,d)
 693
 694 IVEC128_SELECT(32,s32,4,eq,c,d)
 695 IVEC128_SELECT(32,u32,4,eq,c,d)
 696 IVEC128_SELECT(32,32,4,eq,c,d)
 697 IVEC128_SELECT(32,s32,4,neq,c,d)
 698 IVEC128_SELECT(32,u32,4,neq,c,d)
 699 IVEC128_SELECT(32,32,4,neq,c,d)
 700
 701 IVEC128_SELECT(s8,s8,16,gt,c,d)
 702 IVEC128_SELECT(s8,u8,16,gt,c,d)
 703 IVEC128_SELECT(s8,8,16,gt,c,d)
 704 IVEC128_SELECT(s8,s8,16,lt,c,d)
 705 IVEC128_SELECT(s8,u8,16,lt,c,d)
 706 IVEC128_SELECT(s8,8,16,lt,c,d)
 707
 708 IVEC128_SELECT(s16,s16,8,gt,c,d)
 709 IVEC128_SELECT(s16,u16,8,gt,c,d)
 710 IVEC128_SELECT(s16,16,8,gt,c,d)
 711 IVEC128_SELECT(s16,s16,8,lt,c,d)
 712 IVEC128_SELECT(s16,u16,8,lt,c,d)
 713 IVEC128_SELECT(s16,16,8,lt,c,d)
 714
 715 #undef IVEC128_SELECT
 716
 717 class F64vec2
 718 {
 719 protected:
 720   __m128d vec;
 721 public:
 722
 723   F64vec2() {}
 724
 725   F64vec2(__m128d m) { vec = m;}
 726
 727   F64vec2(double d1,double d0) { vec= _mm_set_pd(d1,d0); }
 728
 729   EXPLICIT F64vec2(double d) { vec = _mm_set1_pd(d); }
 730
 731   operator __m128d() const { return vec; }
 732
 733   friend F64vec2 operator &(const F64vec2 &a,const F64vec2 &b) { return _mm_and_pd(a,b); }
 734   friend F64vec2 operator |(const F64vec2 &a,const F64vec2 &b) { return _mm_or_pd(a,b); }
 735   friend F64vec2 operator ^(const F64vec2 &a,const F64vec2 &b) { return _mm_xor_pd(a,b); }
 736
 737   friend F64vec2 operator +(const F64vec2 &a,const F64vec2 &b) { return _mm_add_pd(a,b); }
 738   friend F64vec2 operator -(const F64vec2 &a,const F64vec2 &b) { return _mm_sub_pd(a,b); }
 739   friend F64vec2 operator *(const F64vec2 &a,const F64vec2 &b) { return _mm_mul_pd(a,b); }
 740   friend F64vec2 operator /(const F64vec2 &a,const F64vec2 &b) { return _mm_div_pd(a,b); }
 741
 742   F64vec2& operator +=(F64vec2 &a) { return *this = _mm_add_pd(vec,a); }
 743   F64vec2& operator -=(F64vec2 &a) { return *this = _mm_sub_pd(vec,a); }
 744   F64vec2& operator *=(F64vec2 &a) { return *this = _mm_mul_pd(vec,a); }
 745   F64vec2& operator /=(F64vec2 &a) { return *this = _mm_div_pd(vec,a); }
 746   F64vec2& operator &=(F64vec2 &a) { return *this = _mm_and_pd(vec,a); }
 747   F64vec2& operator |=(F64vec2 &a) { return *this = _mm_or_pd(vec,a); }
 748   F64vec2& operator ^=(F64vec2 &a) { return *this = _mm_xor_pd(vec,a); }
 749
 750   friend double add_horizontal(F64vec2 &a)
 751   {
 752     F64vec2 ftemp = _mm_add_sd(a,_mm_shuffle_pd(a,a,1));
 753     return ftemp[0];
 754   }
 755
 756   friend F64vec2 andnot(const F64vec2 &a,const F64vec2 &b) { return _mm_andnot_pd(a,b); }
 757
 758   friend F64vec2 sqrt(const F64vec2 &a) { return _mm_sqrt_pd(a); }
 759
 760 #define F64vec2_COMP(op) friend F64vec2 cmp##op (const F64vec2 &a,const F64vec2 &b) { return _mm_cmp##op##_pd(a,b); }
 761   F64vec2_COMP(eq)
 762     F64vec2_COMP(lt)
 763     F64vec2_COMP(le)
 764     F64vec2_COMP(gt)
 765     F64vec2_COMP(ge)
 766     F64vec2_COMP(ngt)
 767     F64vec2_COMP(nge)
 768     F64vec2_COMP(neq)
 769     F64vec2_COMP(nlt)
 770     F64vec2_COMP(nle)
 771 #undef F64vec2_COMP
 772
 773     friend F64vec2 simd_min(const F64vec2 &a,const F64vec2 &b) { return _mm_min_pd(a,b); }
 774   friend F64vec2 simd_max(const F64vec2 &a,const F64vec2 &b) { return _mm_max_pd(a,b); }
 775
 776 #define F64vec2_COMI(op) friend int comi##op (const F64vec2 &a,const F64vec2 &b) { return _mm_comi##op##_sd(a,b); }
 777   F64vec2_COMI(eq)
 778     F64vec2_COMI(lt)
 779     F64vec2_COMI(le)
 780     F64vec2_COMI(gt)
 781     F64vec2_COMI(ge)
 782     F64vec2_COMI(neq)
 783 #undef F64vec2_COMI
 784
 785 #define F64vec2_UCOMI(op) friend int ucomi##op (const F64vec2 &a,const F64vec2 &b) { return _mm_ucomi##op##_sd(a,b); }
 786     F64vec2_UCOMI(eq)
 787     F64vec2_UCOMI(lt)
 788     F64vec2_UCOMI(le)
 789     F64vec2_UCOMI(gt)
 790     F64vec2_UCOMI(ge)
 791     F64vec2_UCOMI(neq)
 792 #undef F64vec2_UCOMI
 793
 794 #if defined(_ENABLE_VEC_DEBUG)
 795
 796   friend std::ostream & operator<<(std::ostream & os,const F64vec2 &a) {
 797     double *dp = (double*)&a;
 798     os << " [1]:" << *(dp+1)
 799       << " [0]:" << *dp;
 800     return os;
 801   }
 802 #endif
 803
 804   const double &operator[](int i) const {
 805     assert((0 <= i) && (i <= 1));
 806     double *dp = (double*)&vec;
 807     return *(dp+i);
 808   }
 809
 810   double &operator[](int i) {
 811     assert((0 <= i) && (i <= 1));
 812     double *dp = (double*)&vec;
 813     return *(dp+i);
 814   }
 815 };
 816
 817 inline F64vec2 unpack_low(const F64vec2 &a,const F64vec2 &b) { return _mm_unpacklo_pd(a,b); }
 818 inline F64vec2 unpack_high(const F64vec2 &a,const F64vec2 &b) { return _mm_unpackhi_pd(a,b); }
 819 inline int move_mask(const F64vec2 &a) { return _mm_movemask_pd(a); }
 820 inline void loadu(F64vec2 &a,double *p) { a = _mm_loadu_pd(p); }
 821 inline void storeu(double *p,const F64vec2 &a) { _mm_storeu_pd(p,a); }
 822 inline void store_nta(double *p,F64vec2 &a) { _mm_stream_pd(p,a); }
 823
 824 #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)))); }
 825 F64vec2_SELECT(eq)
 826 F64vec2_SELECT(lt)
 827 F64vec2_SELECT(le)
 828 F64vec2_SELECT(gt)
 829 F64vec2_SELECT(ge)
 830 F64vec2_SELECT(neq)
 831 F64vec2_SELECT(nlt)
 832 F64vec2_SELECT(nle)
 833 #undef F64vec2_SELECT
 834
 835 inline int F64vec2ToInt(const F64vec2 &a) { return _mm_cvttsd_si32(a); }
 836 inline F64vec2 F32vec4ToF64vec2(const F32vec4 &a) { return _mm_cvtps_pd(a); }
 837 inline F32vec4 F64vec2ToF32vec4(const F64vec2 &a) { return _mm_cvtpd_ps(a); }
 838 inline F64vec2 IntToF64vec2(const F64vec2 &a,int b) { return _mm_cvtsi32_sd(a,b); }
 839
 840 #pragma pack(pop)
 841 #pragma pack(pop)
 842 #endif
 843 #endif