xmmintrin.h revision 353358
1193326Sed/*===---- xmmintrin.h - SSE intrinsics -------------------------------------=== 2193326Sed * 3353358Sdim * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4353358Sdim * See https://llvm.org/LICENSE.txt for license information. 5353358Sdim * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6193326Sed * 7193326Sed *===-----------------------------------------------------------------------=== 8193326Sed */ 9296417Sdim 10193326Sed#ifndef __XMMINTRIN_H 11193326Sed#define __XMMINTRIN_H 12193326Sed 13193326Sed#include <mmintrin.h> 14193326Sed 15205408Srdivackytypedef int __v4si __attribute__((__vector_size__(16))); 16193326Sedtypedef float __v4sf __attribute__((__vector_size__(16))); 17353358Sdimtypedef float __m128 __attribute__((__vector_size__(16), __aligned__(16))); 18193326Sed 19353358Sdimtypedef float __m128_u __attribute__((__vector_size__(16), __aligned__(1))); 20353358Sdim 21309124Sdim/* Unsigned types */ 22309124Sdimtypedef unsigned int __v4su __attribute__((__vector_size__(16))); 23309124Sdim 24276479Sdim/* This header should only be included in a hosted environment as it depends on 25276479Sdim * a standard library to provide allocation routines. */ 26218893Sdim#if __STDC_HOSTED__ 27193326Sed#include <mm_malloc.h> 28218893Sdim#endif 29193326Sed 30288943Sdim/* Define the default attributes for the functions in this file. */ 31341825Sdim#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("sse"), __min_vector_width__(128))) 32341825Sdim#define __DEFAULT_FN_ATTRS_MMX __attribute__((__always_inline__, __nodebug__, __target__("mmx,sse"), __min_vector_width__(64))) 33288943Sdim 34341825Sdim/// Adds the 32-bit float values in the low-order bits of the operands. 35309124Sdim/// 36309124Sdim/// \headerfile <x86intrin.h> 37309124Sdim/// 38314564Sdim/// This intrinsic corresponds to the <c> VADDSS / ADDSS </c> instructions. 39309124Sdim/// 40309124Sdim/// \param __a 41309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 42309124Sdim/// The lower 32 bits of this operand are used in the calculation. 43309124Sdim/// \param __b 44309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 45309124Sdim/// The lower 32 bits of this operand are used in the calculation. 46309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the sum 47309124Sdim/// of the lower 32 bits of both operands. The upper 96 bits are copied from 48309124Sdim/// the upper 96 bits of the first source operand. 49288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 50249423Sdim_mm_add_ss(__m128 __a, __m128 __b) 51193326Sed{ 52249423Sdim __a[0] += __b[0]; 53249423Sdim return __a; 54193326Sed} 55193326Sed 56341825Sdim/// Adds two 128-bit vectors of [4 x float], and returns the results of 57309124Sdim/// the addition. 58309124Sdim/// 59309124Sdim/// \headerfile <x86intrin.h> 60309124Sdim/// 61314564Sdim/// This intrinsic corresponds to the <c> VADDPS / ADDPS </c> instructions. 62309124Sdim/// 63309124Sdim/// \param __a 64309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 65309124Sdim/// \param __b 66309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 67309124Sdim/// \returns A 128-bit vector of [4 x float] containing the sums of both 68309124Sdim/// operands. 69288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 70249423Sdim_mm_add_ps(__m128 __a, __m128 __b) 71193326Sed{ 72309124Sdim return (__m128)((__v4sf)__a + (__v4sf)__b); 73193326Sed} 74193326Sed 75341825Sdim/// Subtracts the 32-bit float value in the low-order bits of the second 76309124Sdim/// operand from the corresponding value in the first operand. 77309124Sdim/// 78309124Sdim/// \headerfile <x86intrin.h> 79309124Sdim/// 80314564Sdim/// This intrinsic corresponds to the <c> VSUBSS / SUBSS </c> instructions. 81309124Sdim/// 82309124Sdim/// \param __a 83309124Sdim/// A 128-bit vector of [4 x float] containing the minuend. The lower 32 bits 84309124Sdim/// of this operand are used in the calculation. 85309124Sdim/// \param __b 86309124Sdim/// A 128-bit vector of [4 x float] containing the subtrahend. The lower 32 87309124Sdim/// bits of this operand are used in the calculation. 88309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 89309124Sdim/// difference of the lower 32 bits of both operands. The upper 96 bits are 90309124Sdim/// copied from the upper 96 bits of the first source operand. 91288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 92249423Sdim_mm_sub_ss(__m128 __a, __m128 __b) 93193326Sed{ 94249423Sdim __a[0] -= __b[0]; 95249423Sdim return __a; 96193326Sed} 97193326Sed 98341825Sdim/// Subtracts each of the values of the second operand from the first 99309124Sdim/// operand, both of which are 128-bit vectors of [4 x float] and returns 100309124Sdim/// the results of the subtraction. 101309124Sdim/// 102309124Sdim/// \headerfile <x86intrin.h> 103309124Sdim/// 104314564Sdim/// This intrinsic corresponds to the <c> VSUBPS / SUBPS </c> instructions. 105309124Sdim/// 106309124Sdim/// \param __a 107309124Sdim/// A 128-bit vector of [4 x float] containing the minuend. 108309124Sdim/// \param __b 109309124Sdim/// A 128-bit vector of [4 x float] containing the subtrahend. 110309124Sdim/// \returns A 128-bit vector of [4 x float] containing the differences between 111309124Sdim/// both operands. 112288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 113249423Sdim_mm_sub_ps(__m128 __a, __m128 __b) 114193326Sed{ 115309124Sdim return (__m128)((__v4sf)__a - (__v4sf)__b); 116193326Sed} 117193326Sed 118341825Sdim/// Multiplies two 32-bit float values in the low-order bits of the 119309124Sdim/// operands. 120309124Sdim/// 121309124Sdim/// \headerfile <x86intrin.h> 122309124Sdim/// 123314564Sdim/// This intrinsic corresponds to the <c> VMULSS / MULSS </c> instructions. 124309124Sdim/// 125309124Sdim/// \param __a 126309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 127309124Sdim/// The lower 32 bits of this operand are used in the calculation. 128309124Sdim/// \param __b 129309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 130309124Sdim/// The lower 32 bits of this operand are used in the calculation. 131309124Sdim/// \returns A 128-bit vector of [4 x float] containing the product of the lower 132309124Sdim/// 32 bits of both operands. The upper 96 bits are copied from the upper 96 133309124Sdim/// bits of the first source operand. 134288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 135249423Sdim_mm_mul_ss(__m128 __a, __m128 __b) 136193326Sed{ 137249423Sdim __a[0] *= __b[0]; 138249423Sdim return __a; 139193326Sed} 140193326Sed 141341825Sdim/// Multiplies two 128-bit vectors of [4 x float] and returns the 142309124Sdim/// results of the multiplication. 143309124Sdim/// 144309124Sdim/// \headerfile <x86intrin.h> 145309124Sdim/// 146314564Sdim/// This intrinsic corresponds to the <c> VMULPS / MULPS </c> instructions. 147309124Sdim/// 148309124Sdim/// \param __a 149309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 150309124Sdim/// \param __b 151309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 152309124Sdim/// \returns A 128-bit vector of [4 x float] containing the products of both 153309124Sdim/// operands. 154288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 155249423Sdim_mm_mul_ps(__m128 __a, __m128 __b) 156193326Sed{ 157309124Sdim return (__m128)((__v4sf)__a * (__v4sf)__b); 158193326Sed} 159193326Sed 160341825Sdim/// Divides the value in the low-order 32 bits of the first operand by 161309124Sdim/// the corresponding value in the second operand. 162309124Sdim/// 163309124Sdim/// \headerfile <x86intrin.h> 164309124Sdim/// 165314564Sdim/// This intrinsic corresponds to the <c> VDIVSS / DIVSS </c> instructions. 166309124Sdim/// 167309124Sdim/// \param __a 168309124Sdim/// A 128-bit vector of [4 x float] containing the dividend. The lower 32 169309124Sdim/// bits of this operand are used in the calculation. 170309124Sdim/// \param __b 171309124Sdim/// A 128-bit vector of [4 x float] containing the divisor. The lower 32 bits 172309124Sdim/// of this operand are used in the calculation. 173309124Sdim/// \returns A 128-bit vector of [4 x float] containing the quotients of the 174309124Sdim/// lower 32 bits of both operands. The upper 96 bits are copied from the 175309124Sdim/// upper 96 bits of the first source operand. 176288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 177249423Sdim_mm_div_ss(__m128 __a, __m128 __b) 178193326Sed{ 179249423Sdim __a[0] /= __b[0]; 180249423Sdim return __a; 181193326Sed} 182193326Sed 183341825Sdim/// Divides two 128-bit vectors of [4 x float]. 184309124Sdim/// 185309124Sdim/// \headerfile <x86intrin.h> 186309124Sdim/// 187314564Sdim/// This intrinsic corresponds to the <c> VDIVPS / DIVPS </c> instructions. 188309124Sdim/// 189309124Sdim/// \param __a 190309124Sdim/// A 128-bit vector of [4 x float] containing the dividend. 191309124Sdim/// \param __b 192309124Sdim/// A 128-bit vector of [4 x float] containing the divisor. 193309124Sdim/// \returns A 128-bit vector of [4 x float] containing the quotients of both 194309124Sdim/// operands. 195288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 196249423Sdim_mm_div_ps(__m128 __a, __m128 __b) 197193326Sed{ 198309124Sdim return (__m128)((__v4sf)__a / (__v4sf)__b); 199193326Sed} 200193326Sed 201341825Sdim/// Calculates the square root of the value stored in the low-order bits 202309124Sdim/// of a 128-bit vector of [4 x float]. 203309124Sdim/// 204309124Sdim/// \headerfile <x86intrin.h> 205309124Sdim/// 206314564Sdim/// This intrinsic corresponds to the <c> VSQRTSS / SQRTSS </c> instructions. 207309124Sdim/// 208309124Sdim/// \param __a 209309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 210309124Sdim/// used in the calculation. 211309124Sdim/// \returns A 128-bit vector of [4 x float] containing the square root of the 212309124Sdim/// value in the low-order bits of the operand. 213288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 214249423Sdim_mm_sqrt_ss(__m128 __a) 215193326Sed{ 216341825Sdim return (__m128)__builtin_ia32_sqrtss((__v4sf)__a); 217193326Sed} 218193326Sed 219341825Sdim/// Calculates the square roots of the values stored in a 128-bit vector 220309124Sdim/// of [4 x float]. 221309124Sdim/// 222309124Sdim/// \headerfile <x86intrin.h> 223309124Sdim/// 224314564Sdim/// This intrinsic corresponds to the <c> VSQRTPS / SQRTPS </c> instructions. 225309124Sdim/// 226309124Sdim/// \param __a 227309124Sdim/// A 128-bit vector of [4 x float]. 228309124Sdim/// \returns A 128-bit vector of [4 x float] containing the square roots of the 229309124Sdim/// values in the operand. 230288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 231249423Sdim_mm_sqrt_ps(__m128 __a) 232193326Sed{ 233309124Sdim return __builtin_ia32_sqrtps((__v4sf)__a); 234193326Sed} 235193326Sed 236341825Sdim/// Calculates the approximate reciprocal of the value stored in the 237309124Sdim/// low-order bits of a 128-bit vector of [4 x float]. 238309124Sdim/// 239309124Sdim/// \headerfile <x86intrin.h> 240309124Sdim/// 241314564Sdim/// This intrinsic corresponds to the <c> VRCPSS / RCPSS </c> instructions. 242309124Sdim/// 243309124Sdim/// \param __a 244309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 245309124Sdim/// used in the calculation. 246309124Sdim/// \returns A 128-bit vector of [4 x float] containing the approximate 247309124Sdim/// reciprocal of the value in the low-order bits of the operand. 248288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 249249423Sdim_mm_rcp_ss(__m128 __a) 250193326Sed{ 251341825Sdim return (__m128)__builtin_ia32_rcpss((__v4sf)__a); 252193326Sed} 253193326Sed 254341825Sdim/// Calculates the approximate reciprocals of the values stored in a 255309124Sdim/// 128-bit vector of [4 x float]. 256309124Sdim/// 257309124Sdim/// \headerfile <x86intrin.h> 258309124Sdim/// 259314564Sdim/// This intrinsic corresponds to the <c> VRCPPS / RCPPS </c> instructions. 260309124Sdim/// 261309124Sdim/// \param __a 262309124Sdim/// A 128-bit vector of [4 x float]. 263309124Sdim/// \returns A 128-bit vector of [4 x float] containing the approximate 264309124Sdim/// reciprocals of the values in the operand. 265288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 266249423Sdim_mm_rcp_ps(__m128 __a) 267193326Sed{ 268341825Sdim return (__m128)__builtin_ia32_rcpps((__v4sf)__a); 269193326Sed} 270193326Sed 271341825Sdim/// Calculates the approximate reciprocal of the square root of the value 272309124Sdim/// stored in the low-order bits of a 128-bit vector of [4 x float]. 273309124Sdim/// 274309124Sdim/// \headerfile <x86intrin.h> 275309124Sdim/// 276314564Sdim/// This intrinsic corresponds to the <c> VRSQRTSS / RSQRTSS </c> instructions. 277309124Sdim/// 278309124Sdim/// \param __a 279309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 280309124Sdim/// used in the calculation. 281309124Sdim/// \returns A 128-bit vector of [4 x float] containing the approximate 282309124Sdim/// reciprocal of the square root of the value in the low-order bits of the 283309124Sdim/// operand. 284288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 285249423Sdim_mm_rsqrt_ss(__m128 __a) 286193326Sed{ 287341825Sdim return __builtin_ia32_rsqrtss((__v4sf)__a); 288193326Sed} 289193326Sed 290341825Sdim/// Calculates the approximate reciprocals of the square roots of the 291309124Sdim/// values stored in a 128-bit vector of [4 x float]. 292309124Sdim/// 293309124Sdim/// \headerfile <x86intrin.h> 294309124Sdim/// 295314564Sdim/// This intrinsic corresponds to the <c> VRSQRTPS / RSQRTPS </c> instructions. 296309124Sdim/// 297309124Sdim/// \param __a 298309124Sdim/// A 128-bit vector of [4 x float]. 299309124Sdim/// \returns A 128-bit vector of [4 x float] containing the approximate 300309124Sdim/// reciprocals of the square roots of the values in the operand. 301288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 302249423Sdim_mm_rsqrt_ps(__m128 __a) 303193326Sed{ 304309124Sdim return __builtin_ia32_rsqrtps((__v4sf)__a); 305193326Sed} 306193326Sed 307341825Sdim/// Compares two 32-bit float values in the low-order bits of both 308309124Sdim/// operands and returns the lesser value in the low-order bits of the 309309124Sdim/// vector of [4 x float]. 310309124Sdim/// 311309124Sdim/// \headerfile <x86intrin.h> 312309124Sdim/// 313314564Sdim/// This intrinsic corresponds to the <c> VMINSS / MINSS </c> instructions. 314309124Sdim/// 315309124Sdim/// \param __a 316309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 317309124Sdim/// 32 bits of this operand are used in the comparison. 318309124Sdim/// \param __b 319309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 320309124Sdim/// 32 bits of this operand are used in the comparison. 321309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 322309124Sdim/// minimum value between both operands. The upper 96 bits are copied from 323309124Sdim/// the upper 96 bits of the first source operand. 324288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 325249423Sdim_mm_min_ss(__m128 __a, __m128 __b) 326193326Sed{ 327309124Sdim return __builtin_ia32_minss((__v4sf)__a, (__v4sf)__b); 328193326Sed} 329193326Sed 330341825Sdim/// Compares two 128-bit vectors of [4 x float] and returns the lesser 331314564Sdim/// of each pair of values. 332309124Sdim/// 333309124Sdim/// \headerfile <x86intrin.h> 334309124Sdim/// 335314564Sdim/// This intrinsic corresponds to the <c> VMINPS / MINPS </c> instructions. 336309124Sdim/// 337309124Sdim/// \param __a 338309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. 339309124Sdim/// \param __b 340309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. 341309124Sdim/// \returns A 128-bit vector of [4 x float] containing the minimum values 342309124Sdim/// between both operands. 343288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 344249423Sdim_mm_min_ps(__m128 __a, __m128 __b) 345193326Sed{ 346309124Sdim return __builtin_ia32_minps((__v4sf)__a, (__v4sf)__b); 347193326Sed} 348193326Sed 349341825Sdim/// Compares two 32-bit float values in the low-order bits of both 350314564Sdim/// operands and returns the greater value in the low-order bits of a 128-bit 351314564Sdim/// vector of [4 x float]. 352309124Sdim/// 353309124Sdim/// \headerfile <x86intrin.h> 354309124Sdim/// 355314564Sdim/// This intrinsic corresponds to the <c> VMAXSS / MAXSS </c> instructions. 356309124Sdim/// 357309124Sdim/// \param __a 358309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 359309124Sdim/// 32 bits of this operand are used in the comparison. 360309124Sdim/// \param __b 361309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 362309124Sdim/// 32 bits of this operand are used in the comparison. 363309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 364309124Sdim/// maximum value between both operands. The upper 96 bits are copied from 365309124Sdim/// the upper 96 bits of the first source operand. 366288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 367249423Sdim_mm_max_ss(__m128 __a, __m128 __b) 368193326Sed{ 369309124Sdim return __builtin_ia32_maxss((__v4sf)__a, (__v4sf)__b); 370193326Sed} 371193326Sed 372341825Sdim/// Compares two 128-bit vectors of [4 x float] and returns the greater 373309124Sdim/// of each pair of values. 374309124Sdim/// 375309124Sdim/// \headerfile <x86intrin.h> 376309124Sdim/// 377314564Sdim/// This intrinsic corresponds to the <c> VMAXPS / MAXPS </c> instructions. 378309124Sdim/// 379309124Sdim/// \param __a 380309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. 381309124Sdim/// \param __b 382309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. 383309124Sdim/// \returns A 128-bit vector of [4 x float] containing the maximum values 384309124Sdim/// between both operands. 385288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 386249423Sdim_mm_max_ps(__m128 __a, __m128 __b) 387193326Sed{ 388309124Sdim return __builtin_ia32_maxps((__v4sf)__a, (__v4sf)__b); 389193326Sed} 390193326Sed 391341825Sdim/// Performs a bitwise AND of two 128-bit vectors of [4 x float]. 392309124Sdim/// 393309124Sdim/// \headerfile <x86intrin.h> 394309124Sdim/// 395314564Sdim/// This intrinsic corresponds to the <c> VANDPS / ANDPS </c> instructions. 396309124Sdim/// 397309124Sdim/// \param __a 398309124Sdim/// A 128-bit vector containing one of the source operands. 399309124Sdim/// \param __b 400309124Sdim/// A 128-bit vector containing one of the source operands. 401309124Sdim/// \returns A 128-bit vector of [4 x float] containing the bitwise AND of the 402309124Sdim/// values between both operands. 403288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 404249423Sdim_mm_and_ps(__m128 __a, __m128 __b) 405193326Sed{ 406309124Sdim return (__m128)((__v4su)__a & (__v4su)__b); 407193326Sed} 408193326Sed 409341825Sdim/// Performs a bitwise AND of two 128-bit vectors of [4 x float], using 410309124Sdim/// the one's complement of the values contained in the first source 411309124Sdim/// operand. 412309124Sdim/// 413309124Sdim/// \headerfile <x86intrin.h> 414309124Sdim/// 415314564Sdim/// This intrinsic corresponds to the <c> VANDNPS / ANDNPS </c> instructions. 416309124Sdim/// 417309124Sdim/// \param __a 418309124Sdim/// A 128-bit vector of [4 x float] containing the first source operand. The 419309124Sdim/// one's complement of this value is used in the bitwise AND. 420309124Sdim/// \param __b 421309124Sdim/// A 128-bit vector of [4 x float] containing the second source operand. 422309124Sdim/// \returns A 128-bit vector of [4 x float] containing the bitwise AND of the 423309124Sdim/// one's complement of the first operand and the values in the second 424309124Sdim/// operand. 425288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 426249423Sdim_mm_andnot_ps(__m128 __a, __m128 __b) 427193326Sed{ 428309124Sdim return (__m128)(~(__v4su)__a & (__v4su)__b); 429193326Sed} 430193326Sed 431341825Sdim/// Performs a bitwise OR of two 128-bit vectors of [4 x float]. 432309124Sdim/// 433309124Sdim/// \headerfile <x86intrin.h> 434309124Sdim/// 435314564Sdim/// This intrinsic corresponds to the <c> VORPS / ORPS </c> instructions. 436309124Sdim/// 437309124Sdim/// \param __a 438309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 439309124Sdim/// \param __b 440309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 441309124Sdim/// \returns A 128-bit vector of [4 x float] containing the bitwise OR of the 442309124Sdim/// values between both operands. 443288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 444249423Sdim_mm_or_ps(__m128 __a, __m128 __b) 445193326Sed{ 446309124Sdim return (__m128)((__v4su)__a | (__v4su)__b); 447193326Sed} 448193326Sed 449341825Sdim/// Performs a bitwise exclusive OR of two 128-bit vectors of 450309124Sdim/// [4 x float]. 451309124Sdim/// 452309124Sdim/// \headerfile <x86intrin.h> 453309124Sdim/// 454314564Sdim/// This intrinsic corresponds to the <c> VXORPS / XORPS </c> instructions. 455309124Sdim/// 456309124Sdim/// \param __a 457309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 458309124Sdim/// \param __b 459309124Sdim/// A 128-bit vector of [4 x float] containing one of the source operands. 460309124Sdim/// \returns A 128-bit vector of [4 x float] containing the bitwise exclusive OR 461309124Sdim/// of the values between both operands. 462288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 463249423Sdim_mm_xor_ps(__m128 __a, __m128 __b) 464193326Sed{ 465309124Sdim return (__m128)((__v4su)__a ^ (__v4su)__b); 466193326Sed} 467193326Sed 468341825Sdim/// Compares two 32-bit float values in the low-order bits of both 469309124Sdim/// operands for equality and returns the result of the comparison in the 470309124Sdim/// low-order bits of a vector [4 x float]. 471309124Sdim/// 472309124Sdim/// \headerfile <x86intrin.h> 473309124Sdim/// 474314564Sdim/// This intrinsic corresponds to the <c> VCMPEQSS / CMPEQSS </c> instructions. 475309124Sdim/// 476309124Sdim/// \param __a 477309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 478309124Sdim/// 32 bits of this operand are used in the comparison. 479309124Sdim/// \param __b 480309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 481309124Sdim/// 32 bits of this operand are used in the comparison. 482309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 483309124Sdim/// in the low-order bits. 484288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 485249423Sdim_mm_cmpeq_ss(__m128 __a, __m128 __b) 486193326Sed{ 487309124Sdim return (__m128)__builtin_ia32_cmpeqss((__v4sf)__a, (__v4sf)__b); 488193326Sed} 489193326Sed 490341825Sdim/// Compares each of the corresponding 32-bit float values of the 491309124Sdim/// 128-bit vectors of [4 x float] for equality. 492309124Sdim/// 493309124Sdim/// \headerfile <x86intrin.h> 494309124Sdim/// 495314564Sdim/// This intrinsic corresponds to the <c> VCMPEQPS / CMPEQPS </c> instructions. 496309124Sdim/// 497309124Sdim/// \param __a 498309124Sdim/// A 128-bit vector of [4 x float]. 499309124Sdim/// \param __b 500309124Sdim/// A 128-bit vector of [4 x float]. 501309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 502288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 503249423Sdim_mm_cmpeq_ps(__m128 __a, __m128 __b) 504193326Sed{ 505309124Sdim return (__m128)__builtin_ia32_cmpeqps((__v4sf)__a, (__v4sf)__b); 506193326Sed} 507193326Sed 508341825Sdim/// Compares two 32-bit float values in the low-order bits of both 509309124Sdim/// operands to determine if the value in the first operand is less than the 510309124Sdim/// corresponding value in the second operand and returns the result of the 511309124Sdim/// comparison in the low-order bits of a vector of [4 x float]. 512309124Sdim/// 513309124Sdim/// \headerfile <x86intrin.h> 514309124Sdim/// 515314564Sdim/// This intrinsic corresponds to the <c> VCMPLTSS / CMPLTSS </c> instructions. 516309124Sdim/// 517309124Sdim/// \param __a 518309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 519309124Sdim/// 32 bits of this operand are used in the comparison. 520309124Sdim/// \param __b 521309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 522309124Sdim/// 32 bits of this operand are used in the comparison. 523309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 524309124Sdim/// in the low-order bits. 525288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 526249423Sdim_mm_cmplt_ss(__m128 __a, __m128 __b) 527193326Sed{ 528309124Sdim return (__m128)__builtin_ia32_cmpltss((__v4sf)__a, (__v4sf)__b); 529193326Sed} 530193326Sed 531341825Sdim/// Compares each of the corresponding 32-bit float values of the 532309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 533309124Sdim/// operand are less than those in the second operand. 534309124Sdim/// 535309124Sdim/// \headerfile <x86intrin.h> 536309124Sdim/// 537314564Sdim/// This intrinsic corresponds to the <c> VCMPLTPS / CMPLTPS </c> instructions. 538309124Sdim/// 539309124Sdim/// \param __a 540309124Sdim/// A 128-bit vector of [4 x float]. 541309124Sdim/// \param __b 542309124Sdim/// A 128-bit vector of [4 x float]. 543309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 544288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 545249423Sdim_mm_cmplt_ps(__m128 __a, __m128 __b) 546193326Sed{ 547309124Sdim return (__m128)__builtin_ia32_cmpltps((__v4sf)__a, (__v4sf)__b); 548193326Sed} 549193326Sed 550341825Sdim/// Compares two 32-bit float values in the low-order bits of both 551309124Sdim/// operands to determine if the value in the first operand is less than or 552309124Sdim/// equal to the corresponding value in the second operand and returns the 553309124Sdim/// result of the comparison in the low-order bits of a vector of 554309124Sdim/// [4 x float]. 555309124Sdim/// 556309124Sdim/// \headerfile <x86intrin.h> 557309124Sdim/// 558314564Sdim/// This intrinsic corresponds to the <c> VCMPLESS / CMPLESS </c> instructions. 559309124Sdim/// 560309124Sdim/// \param __a 561309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 562309124Sdim/// 32 bits of this operand are used in the comparison. 563309124Sdim/// \param __b 564309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 565309124Sdim/// 32 bits of this operand are used in the comparison. 566309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 567309124Sdim/// in the low-order bits. 568288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 569249423Sdim_mm_cmple_ss(__m128 __a, __m128 __b) 570193326Sed{ 571309124Sdim return (__m128)__builtin_ia32_cmpless((__v4sf)__a, (__v4sf)__b); 572193326Sed} 573193326Sed 574341825Sdim/// Compares each of the corresponding 32-bit float values of the 575309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 576309124Sdim/// operand are less than or equal to those in the second operand. 577309124Sdim/// 578309124Sdim/// \headerfile <x86intrin.h> 579309124Sdim/// 580314564Sdim/// This intrinsic corresponds to the <c> VCMPLEPS / CMPLEPS </c> instructions. 581309124Sdim/// 582309124Sdim/// \param __a 583309124Sdim/// A 128-bit vector of [4 x float]. 584309124Sdim/// \param __b 585309124Sdim/// A 128-bit vector of [4 x float]. 586309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 587288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 588249423Sdim_mm_cmple_ps(__m128 __a, __m128 __b) 589193326Sed{ 590309124Sdim return (__m128)__builtin_ia32_cmpleps((__v4sf)__a, (__v4sf)__b); 591193326Sed} 592193326Sed 593341825Sdim/// Compares two 32-bit float values in the low-order bits of both 594309124Sdim/// operands to determine if the value in the first operand is greater than 595309124Sdim/// the corresponding value in the second operand and returns the result of 596309124Sdim/// the comparison in the low-order bits of a vector of [4 x float]. 597309124Sdim/// 598309124Sdim/// \headerfile <x86intrin.h> 599309124Sdim/// 600314564Sdim/// This intrinsic corresponds to the <c> VCMPLTSS / CMPLTSS </c> instructions. 601309124Sdim/// 602309124Sdim/// \param __a 603309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 604309124Sdim/// 32 bits of this operand are used in the comparison. 605309124Sdim/// \param __b 606309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 607309124Sdim/// 32 bits of this operand are used in the comparison. 608309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 609309124Sdim/// in the low-order bits. 610288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 611249423Sdim_mm_cmpgt_ss(__m128 __a, __m128 __b) 612193326Sed{ 613309124Sdim return (__m128)__builtin_shufflevector((__v4sf)__a, 614309124Sdim (__v4sf)__builtin_ia32_cmpltss((__v4sf)__b, (__v4sf)__a), 615261991Sdim 4, 1, 2, 3); 616193326Sed} 617193326Sed 618341825Sdim/// Compares each of the corresponding 32-bit float values of the 619309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 620309124Sdim/// operand are greater than those in the second operand. 621309124Sdim/// 622309124Sdim/// \headerfile <x86intrin.h> 623309124Sdim/// 624314564Sdim/// This intrinsic corresponds to the <c> VCMPLTPS / CMPLTPS </c> instructions. 625309124Sdim/// 626309124Sdim/// \param __a 627309124Sdim/// A 128-bit vector of [4 x float]. 628309124Sdim/// \param __b 629309124Sdim/// A 128-bit vector of [4 x float]. 630309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 631288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 632249423Sdim_mm_cmpgt_ps(__m128 __a, __m128 __b) 633193326Sed{ 634309124Sdim return (__m128)__builtin_ia32_cmpltps((__v4sf)__b, (__v4sf)__a); 635193326Sed} 636193326Sed 637341825Sdim/// Compares two 32-bit float values in the low-order bits of both 638309124Sdim/// operands to determine if the value in the first operand is greater than 639309124Sdim/// or equal to the corresponding value in the second operand and returns 640309124Sdim/// the result of the comparison in the low-order bits of a vector of 641309124Sdim/// [4 x float]. 642309124Sdim/// 643309124Sdim/// \headerfile <x86intrin.h> 644309124Sdim/// 645314564Sdim/// This intrinsic corresponds to the <c> VCMPLESS / CMPLESS </c> instructions. 646309124Sdim/// 647309124Sdim/// \param __a 648309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 649309124Sdim/// 32 bits of this operand are used in the comparison. 650309124Sdim/// \param __b 651309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 652309124Sdim/// 32 bits of this operand are used in the comparison. 653309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 654309124Sdim/// in the low-order bits. 655288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 656249423Sdim_mm_cmpge_ss(__m128 __a, __m128 __b) 657193326Sed{ 658309124Sdim return (__m128)__builtin_shufflevector((__v4sf)__a, 659309124Sdim (__v4sf)__builtin_ia32_cmpless((__v4sf)__b, (__v4sf)__a), 660261991Sdim 4, 1, 2, 3); 661193326Sed} 662193326Sed 663341825Sdim/// Compares each of the corresponding 32-bit float values of the 664309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 665309124Sdim/// operand are greater than or equal to those in the second operand. 666309124Sdim/// 667309124Sdim/// \headerfile <x86intrin.h> 668309124Sdim/// 669314564Sdim/// This intrinsic corresponds to the <c> VCMPLEPS / CMPLEPS </c> instructions. 670309124Sdim/// 671309124Sdim/// \param __a 672309124Sdim/// A 128-bit vector of [4 x float]. 673309124Sdim/// \param __b 674309124Sdim/// A 128-bit vector of [4 x float]. 675309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 676288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 677249423Sdim_mm_cmpge_ps(__m128 __a, __m128 __b) 678193326Sed{ 679309124Sdim return (__m128)__builtin_ia32_cmpleps((__v4sf)__b, (__v4sf)__a); 680193326Sed} 681193326Sed 682341825Sdim/// Compares two 32-bit float values in the low-order bits of both 683309124Sdim/// operands for inequality and returns the result of the comparison in the 684309124Sdim/// low-order bits of a vector of [4 x float]. 685309124Sdim/// 686309124Sdim/// \headerfile <x86intrin.h> 687309124Sdim/// 688314564Sdim/// This intrinsic corresponds to the <c> VCMPNEQSS / CMPNEQSS </c> 689314564Sdim/// instructions. 690309124Sdim/// 691309124Sdim/// \param __a 692309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 693309124Sdim/// 32 bits of this operand are used in the comparison. 694309124Sdim/// \param __b 695309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 696309124Sdim/// 32 bits of this operand are used in the comparison. 697309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 698309124Sdim/// in the low-order bits. 699288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 700249423Sdim_mm_cmpneq_ss(__m128 __a, __m128 __b) 701193326Sed{ 702309124Sdim return (__m128)__builtin_ia32_cmpneqss((__v4sf)__a, (__v4sf)__b); 703193326Sed} 704193326Sed 705341825Sdim/// Compares each of the corresponding 32-bit float values of the 706309124Sdim/// 128-bit vectors of [4 x float] for inequality. 707309124Sdim/// 708309124Sdim/// \headerfile <x86intrin.h> 709309124Sdim/// 710314564Sdim/// This intrinsic corresponds to the <c> VCMPNEQPS / CMPNEQPS </c> 711314564Sdim/// instructions. 712309124Sdim/// 713309124Sdim/// \param __a 714309124Sdim/// A 128-bit vector of [4 x float]. 715309124Sdim/// \param __b 716309124Sdim/// A 128-bit vector of [4 x float]. 717309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 718288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 719249423Sdim_mm_cmpneq_ps(__m128 __a, __m128 __b) 720193326Sed{ 721309124Sdim return (__m128)__builtin_ia32_cmpneqps((__v4sf)__a, (__v4sf)__b); 722193326Sed} 723193326Sed 724341825Sdim/// Compares two 32-bit float values in the low-order bits of both 725309124Sdim/// operands to determine if the value in the first operand is not less than 726309124Sdim/// the corresponding value in the second operand and returns the result of 727309124Sdim/// the comparison in the low-order bits of a vector of [4 x float]. 728309124Sdim/// 729309124Sdim/// \headerfile <x86intrin.h> 730309124Sdim/// 731314564Sdim/// This intrinsic corresponds to the <c> VCMPNLTSS / CMPNLTSS </c> 732314564Sdim/// instructions. 733309124Sdim/// 734309124Sdim/// \param __a 735309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 736309124Sdim/// 32 bits of this operand are used in the comparison. 737309124Sdim/// \param __b 738309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 739309124Sdim/// 32 bits of this operand are used in the comparison. 740309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 741309124Sdim/// in the low-order bits. 742288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 743249423Sdim_mm_cmpnlt_ss(__m128 __a, __m128 __b) 744193326Sed{ 745309124Sdim return (__m128)__builtin_ia32_cmpnltss((__v4sf)__a, (__v4sf)__b); 746193326Sed} 747193326Sed 748341825Sdim/// Compares each of the corresponding 32-bit float values of the 749309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 750309124Sdim/// operand are not less than those in the second operand. 751309124Sdim/// 752309124Sdim/// \headerfile <x86intrin.h> 753309124Sdim/// 754314564Sdim/// This intrinsic corresponds to the <c> VCMPNLTPS / CMPNLTPS </c> 755314564Sdim/// instructions. 756309124Sdim/// 757309124Sdim/// \param __a 758309124Sdim/// A 128-bit vector of [4 x float]. 759309124Sdim/// \param __b 760309124Sdim/// A 128-bit vector of [4 x float]. 761309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 762288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 763249423Sdim_mm_cmpnlt_ps(__m128 __a, __m128 __b) 764193326Sed{ 765309124Sdim return (__m128)__builtin_ia32_cmpnltps((__v4sf)__a, (__v4sf)__b); 766193326Sed} 767193326Sed 768341825Sdim/// Compares two 32-bit float values in the low-order bits of both 769309124Sdim/// operands to determine if the value in the first operand is not less than 770309124Sdim/// or equal to the corresponding value in the second operand and returns 771309124Sdim/// the result of the comparison in the low-order bits of a vector of 772309124Sdim/// [4 x float]. 773309124Sdim/// 774309124Sdim/// \headerfile <x86intrin.h> 775309124Sdim/// 776314564Sdim/// This intrinsic corresponds to the <c> VCMPNLESS / CMPNLESS </c> 777314564Sdim/// instructions. 778309124Sdim/// 779309124Sdim/// \param __a 780309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 781309124Sdim/// 32 bits of this operand are used in the comparison. 782309124Sdim/// \param __b 783309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 784309124Sdim/// 32 bits of this operand are used in the comparison. 785309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 786309124Sdim/// in the low-order bits. 787288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 788249423Sdim_mm_cmpnle_ss(__m128 __a, __m128 __b) 789193326Sed{ 790309124Sdim return (__m128)__builtin_ia32_cmpnless((__v4sf)__a, (__v4sf)__b); 791193326Sed} 792193326Sed 793341825Sdim/// Compares each of the corresponding 32-bit float values of the 794309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 795309124Sdim/// operand are not less than or equal to those in the second operand. 796309124Sdim/// 797309124Sdim/// \headerfile <x86intrin.h> 798309124Sdim/// 799314564Sdim/// This intrinsic corresponds to the <c> VCMPNLEPS / CMPNLEPS </c> 800314564Sdim/// instructions. 801309124Sdim/// 802309124Sdim/// \param __a 803309124Sdim/// A 128-bit vector of [4 x float]. 804309124Sdim/// \param __b 805309124Sdim/// A 128-bit vector of [4 x float]. 806309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 807288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 808249423Sdim_mm_cmpnle_ps(__m128 __a, __m128 __b) 809193326Sed{ 810309124Sdim return (__m128)__builtin_ia32_cmpnleps((__v4sf)__a, (__v4sf)__b); 811193326Sed} 812193326Sed 813341825Sdim/// Compares two 32-bit float values in the low-order bits of both 814309124Sdim/// operands to determine if the value in the first operand is not greater 815309124Sdim/// than the corresponding value in the second operand and returns the 816309124Sdim/// result of the comparison in the low-order bits of a vector of 817309124Sdim/// [4 x float]. 818309124Sdim/// 819309124Sdim/// \headerfile <x86intrin.h> 820309124Sdim/// 821314564Sdim/// This intrinsic corresponds to the <c> VCMPNLTSS / CMPNLTSS </c> 822314564Sdim/// instructions. 823309124Sdim/// 824309124Sdim/// \param __a 825309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 826309124Sdim/// 32 bits of this operand are used in the comparison. 827309124Sdim/// \param __b 828309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 829309124Sdim/// 32 bits of this operand are used in the comparison. 830309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 831309124Sdim/// in the low-order bits. 832288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 833249423Sdim_mm_cmpngt_ss(__m128 __a, __m128 __b) 834193326Sed{ 835309124Sdim return (__m128)__builtin_shufflevector((__v4sf)__a, 836309124Sdim (__v4sf)__builtin_ia32_cmpnltss((__v4sf)__b, (__v4sf)__a), 837261991Sdim 4, 1, 2, 3); 838193326Sed} 839193326Sed 840341825Sdim/// Compares each of the corresponding 32-bit float values of the 841309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 842309124Sdim/// operand are not greater than those in the second operand. 843309124Sdim/// 844309124Sdim/// \headerfile <x86intrin.h> 845309124Sdim/// 846314564Sdim/// This intrinsic corresponds to the <c> VCMPNLTPS / CMPNLTPS </c> 847314564Sdim/// instructions. 848309124Sdim/// 849309124Sdim/// \param __a 850309124Sdim/// A 128-bit vector of [4 x float]. 851309124Sdim/// \param __b 852309124Sdim/// A 128-bit vector of [4 x float]. 853309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 854288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 855249423Sdim_mm_cmpngt_ps(__m128 __a, __m128 __b) 856193326Sed{ 857309124Sdim return (__m128)__builtin_ia32_cmpnltps((__v4sf)__b, (__v4sf)__a); 858193326Sed} 859193326Sed 860341825Sdim/// Compares two 32-bit float values in the low-order bits of both 861309124Sdim/// operands to determine if the value in the first operand is not greater 862309124Sdim/// than or equal to the corresponding value in the second operand and 863309124Sdim/// returns the result of the comparison in the low-order bits of a vector 864309124Sdim/// of [4 x float]. 865309124Sdim/// 866309124Sdim/// \headerfile <x86intrin.h> 867309124Sdim/// 868314564Sdim/// This intrinsic corresponds to the <c> VCMPNLESS / CMPNLESS </c> 869314564Sdim/// instructions. 870309124Sdim/// 871309124Sdim/// \param __a 872309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 873309124Sdim/// 32 bits of this operand are used in the comparison. 874309124Sdim/// \param __b 875309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 876309124Sdim/// 32 bits of this operand are used in the comparison. 877309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 878309124Sdim/// in the low-order bits. 879288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 880249423Sdim_mm_cmpnge_ss(__m128 __a, __m128 __b) 881193326Sed{ 882309124Sdim return (__m128)__builtin_shufflevector((__v4sf)__a, 883309124Sdim (__v4sf)__builtin_ia32_cmpnless((__v4sf)__b, (__v4sf)__a), 884261991Sdim 4, 1, 2, 3); 885193326Sed} 886193326Sed 887341825Sdim/// Compares each of the corresponding 32-bit float values of the 888309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 889309124Sdim/// operand are not greater than or equal to those in the second operand. 890309124Sdim/// 891309124Sdim/// \headerfile <x86intrin.h> 892309124Sdim/// 893314564Sdim/// This intrinsic corresponds to the <c> VCMPNLEPS / CMPNLEPS </c> 894314564Sdim/// instructions. 895309124Sdim/// 896309124Sdim/// \param __a 897309124Sdim/// A 128-bit vector of [4 x float]. 898309124Sdim/// \param __b 899309124Sdim/// A 128-bit vector of [4 x float]. 900309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 901288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 902249423Sdim_mm_cmpnge_ps(__m128 __a, __m128 __b) 903193326Sed{ 904309124Sdim return (__m128)__builtin_ia32_cmpnleps((__v4sf)__b, (__v4sf)__a); 905193326Sed} 906193326Sed 907341825Sdim/// Compares two 32-bit float values in the low-order bits of both 908309124Sdim/// operands to determine if the value in the first operand is ordered with 909309124Sdim/// respect to the corresponding value in the second operand and returns the 910309124Sdim/// result of the comparison in the low-order bits of a vector of 911309124Sdim/// [4 x float]. 912309124Sdim/// 913309124Sdim/// \headerfile <x86intrin.h> 914309124Sdim/// 915314564Sdim/// This intrinsic corresponds to the <c> VCMPORDSS / CMPORDSS </c> 916314564Sdim/// instructions. 917309124Sdim/// 918309124Sdim/// \param __a 919309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 920309124Sdim/// 32 bits of this operand are used in the comparison. 921309124Sdim/// \param __b 922309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 923309124Sdim/// 32 bits of this operand are used in the comparison. 924309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 925309124Sdim/// in the low-order bits. 926288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 927249423Sdim_mm_cmpord_ss(__m128 __a, __m128 __b) 928193326Sed{ 929309124Sdim return (__m128)__builtin_ia32_cmpordss((__v4sf)__a, (__v4sf)__b); 930193326Sed} 931193326Sed 932341825Sdim/// Compares each of the corresponding 32-bit float values of the 933309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 934309124Sdim/// operand are ordered with respect to those in the second operand. 935309124Sdim/// 936309124Sdim/// \headerfile <x86intrin.h> 937309124Sdim/// 938314564Sdim/// This intrinsic corresponds to the <c> VCMPORDPS / CMPORDPS </c> 939314564Sdim/// instructions. 940309124Sdim/// 941309124Sdim/// \param __a 942309124Sdim/// A 128-bit vector of [4 x float]. 943309124Sdim/// \param __b 944309124Sdim/// A 128-bit vector of [4 x float]. 945309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 946288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 947249423Sdim_mm_cmpord_ps(__m128 __a, __m128 __b) 948193326Sed{ 949309124Sdim return (__m128)__builtin_ia32_cmpordps((__v4sf)__a, (__v4sf)__b); 950193326Sed} 951193326Sed 952341825Sdim/// Compares two 32-bit float values in the low-order bits of both 953309124Sdim/// operands to determine if the value in the first operand is unordered 954309124Sdim/// with respect to the corresponding value in the second operand and 955309124Sdim/// returns the result of the comparison in the low-order bits of a vector 956309124Sdim/// of [4 x float]. 957309124Sdim/// 958309124Sdim/// \headerfile <x86intrin.h> 959309124Sdim/// 960314564Sdim/// This intrinsic corresponds to the <c> VCMPUNORDSS / CMPUNORDSS </c> 961314564Sdim/// instructions. 962309124Sdim/// 963309124Sdim/// \param __a 964309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 965309124Sdim/// 32 bits of this operand are used in the comparison. 966309124Sdim/// \param __b 967309124Sdim/// A 128-bit vector of [4 x float] containing one of the operands. The lower 968309124Sdim/// 32 bits of this operand are used in the comparison. 969309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results 970309124Sdim/// in the low-order bits. 971288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 972249423Sdim_mm_cmpunord_ss(__m128 __a, __m128 __b) 973193326Sed{ 974309124Sdim return (__m128)__builtin_ia32_cmpunordss((__v4sf)__a, (__v4sf)__b); 975193326Sed} 976193326Sed 977341825Sdim/// Compares each of the corresponding 32-bit float values of the 978309124Sdim/// 128-bit vectors of [4 x float] to determine if the values in the first 979309124Sdim/// operand are unordered with respect to those in the second operand. 980309124Sdim/// 981309124Sdim/// \headerfile <x86intrin.h> 982309124Sdim/// 983314564Sdim/// This intrinsic corresponds to the <c> VCMPUNORDPS / CMPUNORDPS </c> 984314564Sdim/// instructions. 985309124Sdim/// 986309124Sdim/// \param __a 987309124Sdim/// A 128-bit vector of [4 x float]. 988309124Sdim/// \param __b 989309124Sdim/// A 128-bit vector of [4 x float]. 990309124Sdim/// \returns A 128-bit vector of [4 x float] containing the comparison results. 991288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 992249423Sdim_mm_cmpunord_ps(__m128 __a, __m128 __b) 993193326Sed{ 994309124Sdim return (__m128)__builtin_ia32_cmpunordps((__v4sf)__a, (__v4sf)__b); 995193326Sed} 996193326Sed 997341825Sdim/// Compares two 32-bit float values in the low-order bits of both 998309124Sdim/// operands for equality and returns the result of the comparison. 999309124Sdim/// 1000341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1001341825Sdim/// 1002309124Sdim/// \headerfile <x86intrin.h> 1003309124Sdim/// 1004314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> 1005314564Sdim/// instructions. 1006309124Sdim/// 1007309124Sdim/// \param __a 1008309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1009309124Sdim/// used in the comparison. 1010309124Sdim/// \param __b 1011309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1012309124Sdim/// used in the comparison. 1013341825Sdim/// \returns An integer containing the comparison results. If either of the 1014341825Sdim/// two lower 32-bit values is NaN, 0 is returned. 1015288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1016249423Sdim_mm_comieq_ss(__m128 __a, __m128 __b) 1017193326Sed{ 1018309124Sdim return __builtin_ia32_comieq((__v4sf)__a, (__v4sf)__b); 1019193326Sed} 1020193326Sed 1021341825Sdim/// Compares two 32-bit float values in the low-order bits of both 1022309124Sdim/// operands to determine if the first operand is less than the second 1023309124Sdim/// operand and returns the result of the comparison. 1024309124Sdim/// 1025341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1026341825Sdim/// 1027309124Sdim/// \headerfile <x86intrin.h> 1028309124Sdim/// 1029314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> 1030314564Sdim/// instructions. 1031309124Sdim/// 1032309124Sdim/// \param __a 1033309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1034309124Sdim/// used in the comparison. 1035309124Sdim/// \param __b 1036309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1037309124Sdim/// used in the comparison. 1038341825Sdim/// \returns An integer containing the comparison results. If either of the two 1039341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1040288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1041249423Sdim_mm_comilt_ss(__m128 __a, __m128 __b) 1042193326Sed{ 1043309124Sdim return __builtin_ia32_comilt((__v4sf)__a, (__v4sf)__b); 1044193326Sed} 1045193326Sed 1046341825Sdim/// Compares two 32-bit float values in the low-order bits of both 1047309124Sdim/// operands to determine if the first operand is less than or equal to the 1048309124Sdim/// second operand and returns the result of the comparison. 1049309124Sdim/// 1050341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1051341825Sdim/// 1052309124Sdim/// \headerfile <x86intrin.h> 1053309124Sdim/// 1054314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. 1055309124Sdim/// 1056309124Sdim/// \param __a 1057309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1058309124Sdim/// used in the comparison. 1059309124Sdim/// \param __b 1060309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1061309124Sdim/// used in the comparison. 1062341825Sdim/// \returns An integer containing the comparison results. If either of the two 1063341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1064288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1065249423Sdim_mm_comile_ss(__m128 __a, __m128 __b) 1066193326Sed{ 1067309124Sdim return __builtin_ia32_comile((__v4sf)__a, (__v4sf)__b); 1068193326Sed} 1069193326Sed 1070341825Sdim/// Compares two 32-bit float values in the low-order bits of both 1071309124Sdim/// operands to determine if the first operand is greater than the second 1072309124Sdim/// operand and returns the result of the comparison. 1073309124Sdim/// 1074341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1075341825Sdim/// 1076309124Sdim/// \headerfile <x86intrin.h> 1077309124Sdim/// 1078314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. 1079309124Sdim/// 1080309124Sdim/// \param __a 1081309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1082309124Sdim/// used in the comparison. 1083309124Sdim/// \param __b 1084309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1085309124Sdim/// used in the comparison. 1086341825Sdim/// \returns An integer containing the comparison results. If either of the 1087341825Sdim/// two lower 32-bit values is NaN, 0 is returned. 1088288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1089249423Sdim_mm_comigt_ss(__m128 __a, __m128 __b) 1090193326Sed{ 1091309124Sdim return __builtin_ia32_comigt((__v4sf)__a, (__v4sf)__b); 1092193326Sed} 1093193326Sed 1094341825Sdim/// Compares two 32-bit float values in the low-order bits of both 1095309124Sdim/// operands to determine if the first operand is greater than or equal to 1096309124Sdim/// the second operand and returns the result of the comparison. 1097309124Sdim/// 1098341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1099341825Sdim/// 1100309124Sdim/// \headerfile <x86intrin.h> 1101309124Sdim/// 1102314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. 1103309124Sdim/// 1104309124Sdim/// \param __a 1105309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1106309124Sdim/// used in the comparison. 1107309124Sdim/// \param __b 1108309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1109309124Sdim/// used in the comparison. 1110341825Sdim/// \returns An integer containing the comparison results. If either of the two 1111341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1112288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1113249423Sdim_mm_comige_ss(__m128 __a, __m128 __b) 1114193326Sed{ 1115309124Sdim return __builtin_ia32_comige((__v4sf)__a, (__v4sf)__b); 1116193326Sed} 1117193326Sed 1118341825Sdim/// Compares two 32-bit float values in the low-order bits of both 1119309124Sdim/// operands to determine if the first operand is not equal to the second 1120309124Sdim/// operand and returns the result of the comparison. 1121309124Sdim/// 1122341825Sdim/// If either of the two lower 32-bit values is NaN, 1 is returned. 1123341825Sdim/// 1124309124Sdim/// \headerfile <x86intrin.h> 1125309124Sdim/// 1126314564Sdim/// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. 1127309124Sdim/// 1128309124Sdim/// \param __a 1129309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1130309124Sdim/// used in the comparison. 1131309124Sdim/// \param __b 1132309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1133309124Sdim/// used in the comparison. 1134341825Sdim/// \returns An integer containing the comparison results. If either of the 1135341825Sdim/// two lower 32-bit values is NaN, 1 is returned. 1136288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1137249423Sdim_mm_comineq_ss(__m128 __a, __m128 __b) 1138193326Sed{ 1139309124Sdim return __builtin_ia32_comineq((__v4sf)__a, (__v4sf)__b); 1140193326Sed} 1141193326Sed 1142341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1143309124Sdim/// the low-order bits of both operands to determine equality and returns 1144309124Sdim/// the result of the comparison. 1145309124Sdim/// 1146341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1147341825Sdim/// 1148309124Sdim/// \headerfile <x86intrin.h> 1149309124Sdim/// 1150314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1151309124Sdim/// 1152309124Sdim/// \param __a 1153309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1154309124Sdim/// used in the comparison. 1155309124Sdim/// \param __b 1156309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1157309124Sdim/// used in the comparison. 1158341825Sdim/// \returns An integer containing the comparison results. If either of the two 1159341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1160288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1161249423Sdim_mm_ucomieq_ss(__m128 __a, __m128 __b) 1162193326Sed{ 1163309124Sdim return __builtin_ia32_ucomieq((__v4sf)__a, (__v4sf)__b); 1164193326Sed} 1165193326Sed 1166341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1167309124Sdim/// the low-order bits of both operands to determine if the first operand is 1168309124Sdim/// less than the second operand and returns the result of the comparison. 1169309124Sdim/// 1170341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1171341825Sdim/// 1172309124Sdim/// \headerfile <x86intrin.h> 1173309124Sdim/// 1174314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1175309124Sdim/// 1176309124Sdim/// \param __a 1177309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1178309124Sdim/// used in the comparison. 1179309124Sdim/// \param __b 1180309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1181309124Sdim/// used in the comparison. 1182341825Sdim/// \returns An integer containing the comparison results. If either of the two 1183341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1184288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1185249423Sdim_mm_ucomilt_ss(__m128 __a, __m128 __b) 1186193326Sed{ 1187309124Sdim return __builtin_ia32_ucomilt((__v4sf)__a, (__v4sf)__b); 1188193326Sed} 1189193326Sed 1190341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1191314564Sdim/// the low-order bits of both operands to determine if the first operand is 1192314564Sdim/// less than or equal to the second operand and returns the result of the 1193314564Sdim/// comparison. 1194309124Sdim/// 1195341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1196341825Sdim/// 1197309124Sdim/// \headerfile <x86intrin.h> 1198309124Sdim/// 1199314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1200309124Sdim/// 1201309124Sdim/// \param __a 1202309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1203309124Sdim/// used in the comparison. 1204309124Sdim/// \param __b 1205309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1206309124Sdim/// used in the comparison. 1207341825Sdim/// \returns An integer containing the comparison results. If either of the two 1208341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1209288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1210249423Sdim_mm_ucomile_ss(__m128 __a, __m128 __b) 1211193326Sed{ 1212309124Sdim return __builtin_ia32_ucomile((__v4sf)__a, (__v4sf)__b); 1213193326Sed} 1214193326Sed 1215341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1216314564Sdim/// the low-order bits of both operands to determine if the first operand is 1217314564Sdim/// greater than the second operand and returns the result of the 1218309124Sdim/// comparison. 1219309124Sdim/// 1220341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1221341825Sdim/// 1222309124Sdim/// \headerfile <x86intrin.h> 1223309124Sdim/// 1224314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1225309124Sdim/// 1226309124Sdim/// \param __a 1227309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1228309124Sdim/// used in the comparison. 1229309124Sdim/// \param __b 1230309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1231309124Sdim/// used in the comparison. 1232341825Sdim/// \returns An integer containing the comparison results. If either of the two 1233341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1234288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1235249423Sdim_mm_ucomigt_ss(__m128 __a, __m128 __b) 1236193326Sed{ 1237309124Sdim return __builtin_ia32_ucomigt((__v4sf)__a, (__v4sf)__b); 1238193326Sed} 1239193326Sed 1240341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1241309124Sdim/// the low-order bits of both operands to determine if the first operand is 1242309124Sdim/// greater than or equal to the second operand and returns the result of 1243309124Sdim/// the comparison. 1244309124Sdim/// 1245341825Sdim/// If either of the two lower 32-bit values is NaN, 0 is returned. 1246341825Sdim/// 1247309124Sdim/// \headerfile <x86intrin.h> 1248309124Sdim/// 1249314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1250309124Sdim/// 1251309124Sdim/// \param __a 1252309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1253309124Sdim/// used in the comparison. 1254309124Sdim/// \param __b 1255309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1256309124Sdim/// used in the comparison. 1257341825Sdim/// \returns An integer containing the comparison results. If either of the two 1258341825Sdim/// lower 32-bit values is NaN, 0 is returned. 1259288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1260249423Sdim_mm_ucomige_ss(__m128 __a, __m128 __b) 1261193326Sed{ 1262309124Sdim return __builtin_ia32_ucomige((__v4sf)__a, (__v4sf)__b); 1263193326Sed} 1264193326Sed 1265341825Sdim/// Performs an unordered comparison of two 32-bit float values using 1266309124Sdim/// the low-order bits of both operands to determine inequality and returns 1267309124Sdim/// the result of the comparison. 1268309124Sdim/// 1269341825Sdim/// If either of the two lower 32-bit values is NaN, 1 is returned. 1270341825Sdim/// 1271309124Sdim/// \headerfile <x86intrin.h> 1272309124Sdim/// 1273314564Sdim/// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. 1274309124Sdim/// 1275309124Sdim/// \param __a 1276309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1277309124Sdim/// used in the comparison. 1278309124Sdim/// \param __b 1279309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1280309124Sdim/// used in the comparison. 1281341825Sdim/// \returns An integer containing the comparison results. If either of the two 1282341825Sdim/// lower 32-bit values is NaN, 1 is returned. 1283288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1284249423Sdim_mm_ucomineq_ss(__m128 __a, __m128 __b) 1285193326Sed{ 1286309124Sdim return __builtin_ia32_ucomineq((__v4sf)__a, (__v4sf)__b); 1287193326Sed} 1288193326Sed 1289341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1290309124Sdim/// [4 x float] into a 32-bit integer. 1291309124Sdim/// 1292309124Sdim/// \headerfile <x86intrin.h> 1293309124Sdim/// 1294314564Sdim/// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> 1295314564Sdim/// instructions. 1296309124Sdim/// 1297309124Sdim/// \param __a 1298309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1299309124Sdim/// used in the conversion. 1300309124Sdim/// \returns A 32-bit integer containing the converted value. 1301288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1302249423Sdim_mm_cvtss_si32(__m128 __a) 1303193326Sed{ 1304309124Sdim return __builtin_ia32_cvtss2si((__v4sf)__a); 1305193326Sed} 1306193326Sed 1307341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1308309124Sdim/// [4 x float] into a 32-bit integer. 1309309124Sdim/// 1310309124Sdim/// \headerfile <x86intrin.h> 1311309124Sdim/// 1312314564Sdim/// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> 1313314564Sdim/// instructions. 1314309124Sdim/// 1315309124Sdim/// \param __a 1316309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1317309124Sdim/// used in the conversion. 1318309124Sdim/// \returns A 32-bit integer containing the converted value. 1319288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1320249423Sdim_mm_cvt_ss2si(__m128 __a) 1321204643Srdivacky{ 1322249423Sdim return _mm_cvtss_si32(__a); 1323204643Srdivacky} 1324204643Srdivacky 1325193576Sed#ifdef __x86_64__ 1326193576Sed 1327341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1328309124Sdim/// [4 x float] into a 64-bit integer. 1329309124Sdim/// 1330309124Sdim/// \headerfile <x86intrin.h> 1331309124Sdim/// 1332314564Sdim/// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> 1333314564Sdim/// instructions. 1334309124Sdim/// 1335309124Sdim/// \param __a 1336309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1337309124Sdim/// used in the conversion. 1338309124Sdim/// \returns A 64-bit integer containing the converted value. 1339288943Sdimstatic __inline__ long long __DEFAULT_FN_ATTRS 1340249423Sdim_mm_cvtss_si64(__m128 __a) 1341193326Sed{ 1342309124Sdim return __builtin_ia32_cvtss2si64((__v4sf)__a); 1343193326Sed} 1344193326Sed 1345193576Sed#endif 1346193576Sed 1347341825Sdim/// Converts two low-order float values in a 128-bit vector of 1348309124Sdim/// [4 x float] into a 64-bit vector of [2 x i32]. 1349309124Sdim/// 1350309124Sdim/// \headerfile <x86intrin.h> 1351309124Sdim/// 1352314564Sdim/// This intrinsic corresponds to the <c> CVTPS2PI </c> instruction. 1353309124Sdim/// 1354309124Sdim/// \param __a 1355309124Sdim/// A 128-bit vector of [4 x float]. 1356309124Sdim/// \returns A 64-bit integer vector containing the converted values. 1357341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 1358249423Sdim_mm_cvtps_pi32(__m128 __a) 1359193326Sed{ 1360309124Sdim return (__m64)__builtin_ia32_cvtps2pi((__v4sf)__a); 1361193326Sed} 1362193326Sed 1363341825Sdim/// Converts two low-order float values in a 128-bit vector of 1364309124Sdim/// [4 x float] into a 64-bit vector of [2 x i32]. 1365309124Sdim/// 1366309124Sdim/// \headerfile <x86intrin.h> 1367309124Sdim/// 1368314564Sdim/// This intrinsic corresponds to the <c> CVTPS2PI </c> instruction. 1369309124Sdim/// 1370309124Sdim/// \param __a 1371309124Sdim/// A 128-bit vector of [4 x float]. 1372309124Sdim/// \returns A 64-bit integer vector containing the converted values. 1373341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 1374249423Sdim_mm_cvt_ps2pi(__m128 __a) 1375212904Sdim{ 1376249423Sdim return _mm_cvtps_pi32(__a); 1377212904Sdim} 1378212904Sdim 1379341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1380309124Sdim/// [4 x float] into a 32-bit integer, truncating the result when it is 1381309124Sdim/// inexact. 1382309124Sdim/// 1383309124Sdim/// \headerfile <x86intrin.h> 1384309124Sdim/// 1385314564Sdim/// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> 1386314564Sdim/// instructions. 1387309124Sdim/// 1388309124Sdim/// \param __a 1389309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1390309124Sdim/// used in the conversion. 1391309124Sdim/// \returns A 32-bit integer containing the converted value. 1392288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1393249423Sdim_mm_cvttss_si32(__m128 __a) 1394193326Sed{ 1395309124Sdim return __builtin_ia32_cvttss2si((__v4sf)__a); 1396193326Sed} 1397193326Sed 1398341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1399309124Sdim/// [4 x float] into a 32-bit integer, truncating the result when it is 1400309124Sdim/// inexact. 1401309124Sdim/// 1402309124Sdim/// \headerfile <x86intrin.h> 1403309124Sdim/// 1404314564Sdim/// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> 1405314564Sdim/// instructions. 1406309124Sdim/// 1407309124Sdim/// \param __a 1408309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1409309124Sdim/// used in the conversion. 1410309124Sdim/// \returns A 32-bit integer containing the converted value. 1411288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 1412249423Sdim_mm_cvtt_ss2si(__m128 __a) 1413204643Srdivacky{ 1414249423Sdim return _mm_cvttss_si32(__a); 1415204643Srdivacky} 1416204643Srdivacky 1417314564Sdim#ifdef __x86_64__ 1418341825Sdim/// Converts a float value contained in the lower 32 bits of a vector of 1419309124Sdim/// [4 x float] into a 64-bit integer, truncating the result when it is 1420309124Sdim/// inexact. 1421309124Sdim/// 1422309124Sdim/// \headerfile <x86intrin.h> 1423309124Sdim/// 1424314564Sdim/// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> 1425314564Sdim/// instructions. 1426309124Sdim/// 1427309124Sdim/// \param __a 1428309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1429309124Sdim/// used in the conversion. 1430309124Sdim/// \returns A 64-bit integer containing the converted value. 1431288943Sdimstatic __inline__ long long __DEFAULT_FN_ATTRS 1432249423Sdim_mm_cvttss_si64(__m128 __a) 1433193326Sed{ 1434309124Sdim return __builtin_ia32_cvttss2si64((__v4sf)__a); 1435193326Sed} 1436314564Sdim#endif 1437193326Sed 1438341825Sdim/// Converts two low-order float values in a 128-bit vector of 1439309124Sdim/// [4 x float] into a 64-bit vector of [2 x i32], truncating the result 1440309124Sdim/// when it is inexact. 1441309124Sdim/// 1442309124Sdim/// \headerfile <x86intrin.h> 1443309124Sdim/// 1444314564Sdim/// This intrinsic corresponds to the <c> CVTTPS2PI / VTTPS2PI </c> 1445314564Sdim/// instructions. 1446309124Sdim/// 1447309124Sdim/// \param __a 1448309124Sdim/// A 128-bit vector of [4 x float]. 1449309124Sdim/// \returns A 64-bit integer vector containing the converted values. 1450341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 1451249423Sdim_mm_cvttps_pi32(__m128 __a) 1452193326Sed{ 1453309124Sdim return (__m64)__builtin_ia32_cvttps2pi((__v4sf)__a); 1454193326Sed} 1455193326Sed 1456341825Sdim/// Converts two low-order float values in a 128-bit vector of [4 x 1457309124Sdim/// float] into a 64-bit vector of [2 x i32], truncating the result when it 1458309124Sdim/// is inexact. 1459309124Sdim/// 1460309124Sdim/// \headerfile <x86intrin.h> 1461309124Sdim/// 1462314564Sdim/// This intrinsic corresponds to the <c> CVTTPS2PI </c> instruction. 1463309124Sdim/// 1464309124Sdim/// \param __a 1465309124Sdim/// A 128-bit vector of [4 x float]. 1466309124Sdim/// \returns A 64-bit integer vector containing the converted values. 1467341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 1468249423Sdim_mm_cvtt_ps2pi(__m128 __a) 1469212904Sdim{ 1470249423Sdim return _mm_cvttps_pi32(__a); 1471212904Sdim} 1472212904Sdim 1473341825Sdim/// Converts a 32-bit signed integer value into a floating point value 1474309124Sdim/// and writes it to the lower 32 bits of the destination. The remaining 1475309124Sdim/// higher order elements of the destination vector are copied from the 1476309124Sdim/// corresponding elements in the first operand. 1477309124Sdim/// 1478309124Sdim/// \headerfile <x86intrin.h> 1479309124Sdim/// 1480314564Sdim/// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. 1481309124Sdim/// 1482309124Sdim/// \param __a 1483309124Sdim/// A 128-bit vector of [4 x float]. 1484309124Sdim/// \param __b 1485309124Sdim/// A 32-bit signed integer operand containing the value to be converted. 1486309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 1487309124Sdim/// converted value of the second operand. The upper 96 bits are copied from 1488309124Sdim/// the upper 96 bits of the first operand. 1489288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1490249423Sdim_mm_cvtsi32_ss(__m128 __a, int __b) 1491193326Sed{ 1492249423Sdim __a[0] = __b; 1493249423Sdim return __a; 1494193326Sed} 1495193326Sed 1496341825Sdim/// Converts a 32-bit signed integer value into a floating point value 1497309124Sdim/// and writes it to the lower 32 bits of the destination. The remaining 1498309124Sdim/// higher order elements of the destination are copied from the 1499309124Sdim/// corresponding elements in the first operand. 1500309124Sdim/// 1501309124Sdim/// \headerfile <x86intrin.h> 1502309124Sdim/// 1503314564Sdim/// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. 1504309124Sdim/// 1505309124Sdim/// \param __a 1506309124Sdim/// A 128-bit vector of [4 x float]. 1507309124Sdim/// \param __b 1508309124Sdim/// A 32-bit signed integer operand containing the value to be converted. 1509309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 1510309124Sdim/// converted value of the second operand. The upper 96 bits are copied from 1511309124Sdim/// the upper 96 bits of the first operand. 1512288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1513249423Sdim_mm_cvt_si2ss(__m128 __a, int __b) 1514212904Sdim{ 1515249423Sdim return _mm_cvtsi32_ss(__a, __b); 1516212904Sdim} 1517212904Sdim 1518193326Sed#ifdef __x86_64__ 1519193326Sed 1520341825Sdim/// Converts a 64-bit signed integer value into a floating point value 1521309124Sdim/// and writes it to the lower 32 bits of the destination. The remaining 1522309124Sdim/// higher order elements of the destination are copied from the 1523309124Sdim/// corresponding elements in the first operand. 1524309124Sdim/// 1525309124Sdim/// \headerfile <x86intrin.h> 1526309124Sdim/// 1527314564Sdim/// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. 1528309124Sdim/// 1529309124Sdim/// \param __a 1530309124Sdim/// A 128-bit vector of [4 x float]. 1531309124Sdim/// \param __b 1532309124Sdim/// A 64-bit signed integer operand containing the value to be converted. 1533309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the 1534309124Sdim/// converted value of the second operand. The upper 96 bits are copied from 1535309124Sdim/// the upper 96 bits of the first operand. 1536288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1537249423Sdim_mm_cvtsi64_ss(__m128 __a, long long __b) 1538193326Sed{ 1539249423Sdim __a[0] = __b; 1540249423Sdim return __a; 1541193326Sed} 1542193326Sed 1543193326Sed#endif 1544193326Sed 1545341825Sdim/// Converts two elements of a 64-bit vector of [2 x i32] into two 1546309124Sdim/// floating point values and writes them to the lower 64-bits of the 1547309124Sdim/// destination. The remaining higher order elements of the destination are 1548309124Sdim/// copied from the corresponding elements in the first operand. 1549309124Sdim/// 1550309124Sdim/// \headerfile <x86intrin.h> 1551309124Sdim/// 1552314564Sdim/// This intrinsic corresponds to the <c> CVTPI2PS </c> instruction. 1553309124Sdim/// 1554309124Sdim/// \param __a 1555309124Sdim/// A 128-bit vector of [4 x float]. 1556309124Sdim/// \param __b 1557309124Sdim/// A 64-bit vector of [2 x i32]. The elements in this vector are converted 1558309124Sdim/// and written to the corresponding low-order elements in the destination. 1559309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the 1560309124Sdim/// converted value of the second operand. The upper 64 bits are copied from 1561309124Sdim/// the upper 64 bits of the first operand. 1562341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 1563249423Sdim_mm_cvtpi32_ps(__m128 __a, __m64 __b) 1564193326Sed{ 1565309124Sdim return __builtin_ia32_cvtpi2ps((__v4sf)__a, (__v2si)__b); 1566193326Sed} 1567193326Sed 1568341825Sdim/// Converts two elements of a 64-bit vector of [2 x i32] into two 1569309124Sdim/// floating point values and writes them to the lower 64-bits of the 1570309124Sdim/// destination. The remaining higher order elements of the destination are 1571309124Sdim/// copied from the corresponding elements in the first operand. 1572309124Sdim/// 1573309124Sdim/// \headerfile <x86intrin.h> 1574309124Sdim/// 1575314564Sdim/// This intrinsic corresponds to the <c> CVTPI2PS </c> instruction. 1576309124Sdim/// 1577309124Sdim/// \param __a 1578309124Sdim/// A 128-bit vector of [4 x float]. 1579309124Sdim/// \param __b 1580309124Sdim/// A 64-bit vector of [2 x i32]. The elements in this vector are converted 1581309124Sdim/// and written to the corresponding low-order elements in the destination. 1582309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the 1583309124Sdim/// converted value from the second operand. The upper 64 bits are copied 1584309124Sdim/// from the upper 64 bits of the first operand. 1585341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 1586249423Sdim_mm_cvt_pi2ps(__m128 __a, __m64 __b) 1587212904Sdim{ 1588249423Sdim return _mm_cvtpi32_ps(__a, __b); 1589212904Sdim} 1590212904Sdim 1591341825Sdim/// Extracts a float value contained in the lower 32 bits of a vector of 1592309124Sdim/// [4 x float]. 1593309124Sdim/// 1594309124Sdim/// \headerfile <x86intrin.h> 1595309124Sdim/// 1596341825Sdim/// This intrinsic has no corresponding instruction. 1597309124Sdim/// 1598309124Sdim/// \param __a 1599309124Sdim/// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are 1600309124Sdim/// used in the extraction. 1601309124Sdim/// \returns A 32-bit float containing the extracted value. 1602288943Sdimstatic __inline__ float __DEFAULT_FN_ATTRS 1603249423Sdim_mm_cvtss_f32(__m128 __a) 1604193326Sed{ 1605249423Sdim return __a[0]; 1606193326Sed} 1607193326Sed 1608341825Sdim/// Loads two packed float values from the address \a __p into the 1609309124Sdim/// high-order bits of a 128-bit vector of [4 x float]. The low-order bits 1610309124Sdim/// are copied from the low-order bits of the first operand. 1611309124Sdim/// 1612309124Sdim/// \headerfile <x86intrin.h> 1613309124Sdim/// 1614314564Sdim/// This intrinsic corresponds to the <c> VMOVHPD / MOVHPD </c> instruction. 1615309124Sdim/// 1616309124Sdim/// \param __a 1617309124Sdim/// A 128-bit vector of [4 x float]. Bits [63:0] are written to bits [63:0] 1618309124Sdim/// of the destination. 1619309124Sdim/// \param __p 1620309124Sdim/// A pointer to two packed float values. Bits [63:0] are written to bits 1621309124Sdim/// [127:64] of the destination. 1622309124Sdim/// \returns A 128-bit vector of [4 x float] containing the moved values. 1623288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1624249423Sdim_mm_loadh_pi(__m128 __a, const __m64 *__p) 1625193326Sed{ 1626226633Sdim typedef float __mm_loadh_pi_v2f32 __attribute__((__vector_size__(8))); 1627226633Sdim struct __mm_loadh_pi_struct { 1628249423Sdim __mm_loadh_pi_v2f32 __u; 1629226633Sdim } __attribute__((__packed__, __may_alias__)); 1630249423Sdim __mm_loadh_pi_v2f32 __b = ((struct __mm_loadh_pi_struct*)__p)->__u; 1631249423Sdim __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1); 1632249423Sdim return __builtin_shufflevector(__a, __bb, 0, 1, 4, 5); 1633193326Sed} 1634193326Sed 1635341825Sdim/// Loads two packed float values from the address \a __p into the 1636314564Sdim/// low-order bits of a 128-bit vector of [4 x float]. The high-order bits 1637314564Sdim/// are copied from the high-order bits of the first operand. 1638309124Sdim/// 1639309124Sdim/// \headerfile <x86intrin.h> 1640309124Sdim/// 1641314564Sdim/// This intrinsic corresponds to the <c> VMOVLPD / MOVLPD </c> instruction. 1642309124Sdim/// 1643309124Sdim/// \param __a 1644309124Sdim/// A 128-bit vector of [4 x float]. Bits [127:64] are written to bits 1645309124Sdim/// [127:64] of the destination. 1646309124Sdim/// \param __p 1647309124Sdim/// A pointer to two packed float values. Bits [63:0] are written to bits 1648309124Sdim/// [63:0] of the destination. 1649309124Sdim/// \returns A 128-bit vector of [4 x float] containing the moved values. 1650288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1651249423Sdim_mm_loadl_pi(__m128 __a, const __m64 *__p) 1652193326Sed{ 1653226633Sdim typedef float __mm_loadl_pi_v2f32 __attribute__((__vector_size__(8))); 1654226633Sdim struct __mm_loadl_pi_struct { 1655249423Sdim __mm_loadl_pi_v2f32 __u; 1656226633Sdim } __attribute__((__packed__, __may_alias__)); 1657249423Sdim __mm_loadl_pi_v2f32 __b = ((struct __mm_loadl_pi_struct*)__p)->__u; 1658249423Sdim __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1); 1659249423Sdim return __builtin_shufflevector(__a, __bb, 4, 5, 2, 3); 1660193326Sed} 1661193326Sed 1662341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float]. The lower 1663309124Sdim/// 32 bits of the vector are initialized with the single-precision 1664309124Sdim/// floating-point value loaded from a specified memory location. The upper 1665309124Sdim/// 96 bits are set to zero. 1666309124Sdim/// 1667309124Sdim/// \headerfile <x86intrin.h> 1668309124Sdim/// 1669314564Sdim/// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. 1670309124Sdim/// 1671309124Sdim/// \param __p 1672309124Sdim/// A pointer to a 32-bit memory location containing a single-precision 1673309124Sdim/// floating-point value. 1674309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. The 1675309124Sdim/// lower 32 bits contain the value loaded from the memory location. The 1676309124Sdim/// upper 96 bits are set to zero. 1677288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1678249423Sdim_mm_load_ss(const float *__p) 1679193326Sed{ 1680226633Sdim struct __mm_load_ss_struct { 1681249423Sdim float __u; 1682226633Sdim } __attribute__((__packed__, __may_alias__)); 1683249423Sdim float __u = ((struct __mm_load_ss_struct*)__p)->__u; 1684341825Sdim return __extension__ (__m128){ __u, 0, 0, 0 }; 1685193326Sed} 1686193326Sed 1687341825Sdim/// Loads a 32-bit float value and duplicates it to all four vector 1688309124Sdim/// elements of a 128-bit vector of [4 x float]. 1689309124Sdim/// 1690309124Sdim/// \headerfile <x86intrin.h> 1691309124Sdim/// 1692341825Sdim/// This intrinsic corresponds to the <c> VBROADCASTSS / MOVSS + shuffling </c> 1693309124Sdim/// instruction. 1694309124Sdim/// 1695309124Sdim/// \param __p 1696309124Sdim/// A pointer to a float value to be loaded and duplicated. 1697314564Sdim/// \returns A 128-bit vector of [4 x float] containing the loaded and 1698314564Sdim/// duplicated values. 1699288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1700249423Sdim_mm_load1_ps(const float *__p) 1701193326Sed{ 1702226633Sdim struct __mm_load1_ps_struct { 1703249423Sdim float __u; 1704226633Sdim } __attribute__((__packed__, __may_alias__)); 1705249423Sdim float __u = ((struct __mm_load1_ps_struct*)__p)->__u; 1706341825Sdim return __extension__ (__m128){ __u, __u, __u, __u }; 1707193326Sed} 1708193326Sed 1709193326Sed#define _mm_load_ps1(p) _mm_load1_ps(p) 1710193326Sed 1711341825Sdim/// Loads a 128-bit floating-point vector of [4 x float] from an aligned 1712309124Sdim/// memory location. 1713309124Sdim/// 1714309124Sdim/// \headerfile <x86intrin.h> 1715309124Sdim/// 1716314564Sdim/// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS </c> instruction. 1717309124Sdim/// 1718309124Sdim/// \param __p 1719309124Sdim/// A pointer to a 128-bit memory location. The address of the memory 1720309124Sdim/// location has to be 128-bit aligned. 1721341825Sdim/// \returns A 128-bit vector of [4 x float] containing the loaded values. 1722288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1723249423Sdim_mm_load_ps(const float *__p) 1724193326Sed{ 1725249423Sdim return *(__m128*)__p; 1726193326Sed} 1727193326Sed 1728341825Sdim/// Loads a 128-bit floating-point vector of [4 x float] from an 1729309124Sdim/// unaligned memory location. 1730309124Sdim/// 1731309124Sdim/// \headerfile <x86intrin.h> 1732309124Sdim/// 1733314564Sdim/// This intrinsic corresponds to the <c> VMOVUPS / MOVUPS </c> instruction. 1734309124Sdim/// 1735309124Sdim/// \param __p 1736309124Sdim/// A pointer to a 128-bit memory location. The address of the memory 1737309124Sdim/// location does not have to be aligned. 1738309124Sdim/// \returns A 128-bit vector of [4 x float] containing the loaded values. 1739288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1740249423Sdim_mm_loadu_ps(const float *__p) 1741193326Sed{ 1742223017Sdim struct __loadu_ps { 1743353358Sdim __m128_u __v; 1744226633Sdim } __attribute__((__packed__, __may_alias__)); 1745249423Sdim return ((struct __loadu_ps*)__p)->__v; 1746193326Sed} 1747193326Sed 1748341825Sdim/// Loads four packed float values, in reverse order, from an aligned 1749309124Sdim/// memory location to 32-bit elements in a 128-bit vector of [4 x float]. 1750309124Sdim/// 1751309124Sdim/// \headerfile <x86intrin.h> 1752309124Sdim/// 1753314564Sdim/// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS + shuffling </c> 1754309124Sdim/// instruction. 1755309124Sdim/// 1756309124Sdim/// \param __p 1757309124Sdim/// A pointer to a 128-bit memory location. The address of the memory 1758309124Sdim/// location has to be 128-bit aligned. 1759309124Sdim/// \returns A 128-bit vector of [4 x float] containing the moved values, loaded 1760309124Sdim/// in reverse order. 1761288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1762249423Sdim_mm_loadr_ps(const float *__p) 1763193326Sed{ 1764249423Sdim __m128 __a = _mm_load_ps(__p); 1765309124Sdim return __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 3, 2, 1, 0); 1766193326Sed} 1767193326Sed 1768341825Sdim/// Create a 128-bit vector of [4 x float] with undefined values. 1769309124Sdim/// 1770309124Sdim/// \headerfile <x86intrin.h> 1771309124Sdim/// 1772309124Sdim/// This intrinsic has no corresponding instruction. 1773309124Sdim/// 1774309124Sdim/// \returns A 128-bit vector of [4 x float] containing undefined values. 1775288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1776309124Sdim_mm_undefined_ps(void) 1777296417Sdim{ 1778296417Sdim return (__m128)__builtin_ia32_undef128(); 1779296417Sdim} 1780296417Sdim 1781341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float]. The lower 1782309124Sdim/// 32 bits of the vector are initialized with the specified single-precision 1783309124Sdim/// floating-point value. The upper 96 bits are set to zero. 1784309124Sdim/// 1785309124Sdim/// \headerfile <x86intrin.h> 1786309124Sdim/// 1787314564Sdim/// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. 1788309124Sdim/// 1789309124Sdim/// \param __w 1790309124Sdim/// A single-precision floating-point value used to initialize the lower 32 1791309124Sdim/// bits of the result. 1792309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. The 1793309124Sdim/// lower 32 bits contain the value provided in the source operand. The 1794309124Sdim/// upper 96 bits are set to zero. 1795296417Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1796249423Sdim_mm_set_ss(float __w) 1797193326Sed{ 1798341825Sdim return __extension__ (__m128){ __w, 0, 0, 0 }; 1799193326Sed} 1800193326Sed 1801341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float], with each 1802309124Sdim/// of the four single-precision floating-point vector elements set to the 1803309124Sdim/// specified single-precision floating-point value. 1804309124Sdim/// 1805309124Sdim/// \headerfile <x86intrin.h> 1806309124Sdim/// 1807314564Sdim/// This intrinsic corresponds to the <c> VPERMILPS / PERMILPS </c> instruction. 1808309124Sdim/// 1809309124Sdim/// \param __w 1810309124Sdim/// A single-precision floating-point value used to initialize each vector 1811309124Sdim/// element of the result. 1812309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. 1813288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1814249423Sdim_mm_set1_ps(float __w) 1815193326Sed{ 1816341825Sdim return __extension__ (__m128){ __w, __w, __w, __w }; 1817193326Sed} 1818193326Sed 1819276479Sdim/* Microsoft specific. */ 1820341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float], with each 1821309124Sdim/// of the four single-precision floating-point vector elements set to the 1822309124Sdim/// specified single-precision floating-point value. 1823309124Sdim/// 1824309124Sdim/// \headerfile <x86intrin.h> 1825309124Sdim/// 1826314564Sdim/// This intrinsic corresponds to the <c> VPERMILPS / PERMILPS </c> instruction. 1827309124Sdim/// 1828309124Sdim/// \param __w 1829309124Sdim/// A single-precision floating-point value used to initialize each vector 1830309124Sdim/// element of the result. 1831309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. 1832288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1833249423Sdim_mm_set_ps1(float __w) 1834193326Sed{ 1835249423Sdim return _mm_set1_ps(__w); 1836193326Sed} 1837193326Sed 1838341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float] 1839309124Sdim/// initialized with the specified single-precision floating-point values. 1840309124Sdim/// 1841309124Sdim/// \headerfile <x86intrin.h> 1842309124Sdim/// 1843309124Sdim/// This intrinsic is a utility function and does not correspond to a specific 1844309124Sdim/// instruction. 1845309124Sdim/// 1846309124Sdim/// \param __z 1847309124Sdim/// A single-precision floating-point value used to initialize bits [127:96] 1848309124Sdim/// of the result. 1849309124Sdim/// \param __y 1850309124Sdim/// A single-precision floating-point value used to initialize bits [95:64] 1851309124Sdim/// of the result. 1852309124Sdim/// \param __x 1853309124Sdim/// A single-precision floating-point value used to initialize bits [63:32] 1854309124Sdim/// of the result. 1855309124Sdim/// \param __w 1856309124Sdim/// A single-precision floating-point value used to initialize bits [31:0] 1857309124Sdim/// of the result. 1858309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. 1859288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1860249423Sdim_mm_set_ps(float __z, float __y, float __x, float __w) 1861193326Sed{ 1862341825Sdim return __extension__ (__m128){ __w, __x, __y, __z }; 1863193326Sed} 1864193326Sed 1865341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float], 1866309124Sdim/// initialized in reverse order with the specified 32-bit single-precision 1867309124Sdim/// float-point values. 1868309124Sdim/// 1869309124Sdim/// \headerfile <x86intrin.h> 1870309124Sdim/// 1871309124Sdim/// This intrinsic is a utility function and does not correspond to a specific 1872309124Sdim/// instruction. 1873309124Sdim/// 1874309124Sdim/// \param __z 1875309124Sdim/// A single-precision floating-point value used to initialize bits [31:0] 1876309124Sdim/// of the result. 1877309124Sdim/// \param __y 1878309124Sdim/// A single-precision floating-point value used to initialize bits [63:32] 1879309124Sdim/// of the result. 1880309124Sdim/// \param __x 1881309124Sdim/// A single-precision floating-point value used to initialize bits [95:64] 1882309124Sdim/// of the result. 1883309124Sdim/// \param __w 1884309124Sdim/// A single-precision floating-point value used to initialize bits [127:96] 1885309124Sdim/// of the result. 1886309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float]. 1887288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1888249423Sdim_mm_setr_ps(float __z, float __y, float __x, float __w) 1889193326Sed{ 1890341825Sdim return __extension__ (__m128){ __z, __y, __x, __w }; 1891193326Sed} 1892193326Sed 1893341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float] initialized 1894309124Sdim/// to zero. 1895309124Sdim/// 1896309124Sdim/// \headerfile <x86intrin.h> 1897309124Sdim/// 1898314564Sdim/// This intrinsic corresponds to the <c> VXORPS / XORPS </c> instruction. 1899309124Sdim/// 1900309124Sdim/// \returns An initialized 128-bit floating-point vector of [4 x float] with 1901309124Sdim/// all elements set to zero. 1902288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 1903193326Sed_mm_setzero_ps(void) 1904193326Sed{ 1905341825Sdim return __extension__ (__m128){ 0, 0, 0, 0 }; 1906193326Sed} 1907193326Sed 1908341825Sdim/// Stores the upper 64 bits of a 128-bit vector of [4 x float] to a 1909309124Sdim/// memory location. 1910309124Sdim/// 1911309124Sdim/// \headerfile <x86intrin.h> 1912309124Sdim/// 1913341825Sdim/// This intrinsic corresponds to the <c> VPEXTRQ / PEXTRQ </c> instruction. 1914309124Sdim/// 1915309124Sdim/// \param __p 1916309124Sdim/// A pointer to a 64-bit memory location. 1917309124Sdim/// \param __a 1918309124Sdim/// A 128-bit vector of [4 x float] containing the values to be stored. 1919288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 1920249423Sdim_mm_storeh_pi(__m64 *__p, __m128 __a) 1921193326Sed{ 1922353358Sdim typedef float __mm_storeh_pi_v2f32 __attribute__((__vector_size__(8))); 1923353358Sdim struct __mm_storeh_pi_struct { 1924353358Sdim __mm_storeh_pi_v2f32 __u; 1925353358Sdim } __attribute__((__packed__, __may_alias__)); 1926353358Sdim ((struct __mm_storeh_pi_struct*)__p)->__u = __builtin_shufflevector(__a, __a, 2, 3); 1927193326Sed} 1928193326Sed 1929341825Sdim/// Stores the lower 64 bits of a 128-bit vector of [4 x float] to a 1930309124Sdim/// memory location. 1931309124Sdim/// 1932309124Sdim/// \headerfile <x86intrin.h> 1933309124Sdim/// 1934314564Sdim/// This intrinsic corresponds to the <c> VMOVLPS / MOVLPS </c> instruction. 1935309124Sdim/// 1936309124Sdim/// \param __p 1937309124Sdim/// A pointer to a memory location that will receive the float values. 1938309124Sdim/// \param __a 1939309124Sdim/// A 128-bit vector of [4 x float] containing the values to be stored. 1940288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 1941249423Sdim_mm_storel_pi(__m64 *__p, __m128 __a) 1942193326Sed{ 1943353358Sdim typedef float __mm_storeh_pi_v2f32 __attribute__((__vector_size__(8))); 1944353358Sdim struct __mm_storeh_pi_struct { 1945353358Sdim __mm_storeh_pi_v2f32 __u; 1946353358Sdim } __attribute__((__packed__, __may_alias__)); 1947353358Sdim ((struct __mm_storeh_pi_struct*)__p)->__u = __builtin_shufflevector(__a, __a, 0, 1); 1948193326Sed} 1949193326Sed 1950341825Sdim/// Stores the lower 32 bits of a 128-bit vector of [4 x float] to a 1951309124Sdim/// memory location. 1952309124Sdim/// 1953309124Sdim/// \headerfile <x86intrin.h> 1954309124Sdim/// 1955314564Sdim/// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. 1956309124Sdim/// 1957309124Sdim/// \param __p 1958309124Sdim/// A pointer to a 32-bit memory location. 1959309124Sdim/// \param __a 1960309124Sdim/// A 128-bit vector of [4 x float] containing the value to be stored. 1961288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 1962249423Sdim_mm_store_ss(float *__p, __m128 __a) 1963193326Sed{ 1964226633Sdim struct __mm_store_ss_struct { 1965249423Sdim float __u; 1966226633Sdim } __attribute__((__packed__, __may_alias__)); 1967249423Sdim ((struct __mm_store_ss_struct*)__p)->__u = __a[0]; 1968193326Sed} 1969193326Sed 1970341825Sdim/// Stores a 128-bit vector of [4 x float] to an unaligned memory 1971314564Sdim/// location. 1972309124Sdim/// 1973309124Sdim/// \headerfile <x86intrin.h> 1974309124Sdim/// 1975314564Sdim/// This intrinsic corresponds to the <c> VMOVUPS / MOVUPS </c> instruction. 1976309124Sdim/// 1977309124Sdim/// \param __p 1978309124Sdim/// A pointer to a 128-bit memory location. The address of the memory 1979309124Sdim/// location does not have to be aligned. 1980309124Sdim/// \param __a 1981309124Sdim/// A 128-bit vector of [4 x float] containing the values to be stored. 1982288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 1983249423Sdim_mm_storeu_ps(float *__p, __m128 __a) 1984193326Sed{ 1985309124Sdim struct __storeu_ps { 1986353358Sdim __m128_u __v; 1987309124Sdim } __attribute__((__packed__, __may_alias__)); 1988309124Sdim ((struct __storeu_ps*)__p)->__v = __a; 1989193326Sed} 1990193326Sed 1991341825Sdim/// Stores a 128-bit vector of [4 x float] into an aligned memory 1992314564Sdim/// location. 1993309124Sdim/// 1994309124Sdim/// \headerfile <x86intrin.h> 1995309124Sdim/// 1996314564Sdim/// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS </c> instruction. 1997309124Sdim/// 1998309124Sdim/// \param __p 1999314564Sdim/// A pointer to a 128-bit memory location. The address of the memory 2000314564Sdim/// location has to be 16-byte aligned. 2001309124Sdim/// \param __a 2002314564Sdim/// A 128-bit vector of [4 x float] containing the values to be stored. 2003288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 2004309124Sdim_mm_store_ps(float *__p, __m128 __a) 2005193326Sed{ 2006309124Sdim *(__m128*)__p = __a; 2007193326Sed} 2008193326Sed 2009341825Sdim/// Stores the lower 32 bits of a 128-bit vector of [4 x float] into 2010309124Sdim/// four contiguous elements in an aligned memory location. 2011309124Sdim/// 2012309124Sdim/// \headerfile <x86intrin.h> 2013309124Sdim/// 2014314564Sdim/// This intrinsic corresponds to <c> VMOVAPS / MOVAPS + shuffling </c> 2015309124Sdim/// instruction. 2016309124Sdim/// 2017309124Sdim/// \param __p 2018309124Sdim/// A pointer to a 128-bit memory location. 2019309124Sdim/// \param __a 2020309124Sdim/// A 128-bit vector of [4 x float] whose lower 32 bits are stored to each 2021314564Sdim/// of the four contiguous elements pointed by \a __p. 2022288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 2023309124Sdim_mm_store1_ps(float *__p, __m128 __a) 2024212904Sdim{ 2025309124Sdim __a = __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 0, 0, 0, 0); 2026309124Sdim _mm_store_ps(__p, __a); 2027212904Sdim} 2028212904Sdim 2029341825Sdim/// Stores the lower 32 bits of a 128-bit vector of [4 x float] into 2030314564Sdim/// four contiguous elements in an aligned memory location. 2031309124Sdim/// 2032309124Sdim/// \headerfile <x86intrin.h> 2033309124Sdim/// 2034314564Sdim/// This intrinsic corresponds to <c> VMOVAPS / MOVAPS + shuffling </c> 2035314564Sdim/// instruction. 2036309124Sdim/// 2037309124Sdim/// \param __p 2038314564Sdim/// A pointer to a 128-bit memory location. 2039309124Sdim/// \param __a 2040314564Sdim/// A 128-bit vector of [4 x float] whose lower 32 bits are stored to each 2041314564Sdim/// of the four contiguous elements pointed by \a __p. 2042288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 2043309124Sdim_mm_store_ps1(float *__p, __m128 __a) 2044193326Sed{ 2045341825Sdim _mm_store1_ps(__p, __a); 2046193326Sed} 2047193326Sed 2048341825Sdim/// Stores float values from a 128-bit vector of [4 x float] to an 2049309124Sdim/// aligned memory location in reverse order. 2050309124Sdim/// 2051309124Sdim/// \headerfile <x86intrin.h> 2052309124Sdim/// 2053314564Sdim/// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS + shuffling </c> 2054309124Sdim/// instruction. 2055309124Sdim/// 2056309124Sdim/// \param __p 2057309124Sdim/// A pointer to a 128-bit memory location. The address of the memory 2058309124Sdim/// location has to be 128-bit aligned. 2059309124Sdim/// \param __a 2060309124Sdim/// A 128-bit vector of [4 x float] containing the values to be stored. 2061288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 2062249423Sdim_mm_storer_ps(float *__p, __m128 __a) 2063193326Sed{ 2064309124Sdim __a = __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 3, 2, 1, 0); 2065249423Sdim _mm_store_ps(__p, __a); 2066193326Sed} 2067193326Sed 2068327952Sdim#define _MM_HINT_ET0 7 2069327952Sdim#define _MM_HINT_ET1 6 2070327952Sdim#define _MM_HINT_T0 3 2071327952Sdim#define _MM_HINT_T1 2 2072327952Sdim#define _MM_HINT_T2 1 2073193326Sed#define _MM_HINT_NTA 0 2074193326Sed 2075276479Sdim#ifndef _MSC_VER 2076210299Sed/* FIXME: We have to #define this because "sel" must be a constant integer, and 2077193326Sed Sema doesn't do any form of constant propagation yet. */ 2078193326Sed 2079341825Sdim/// Loads one cache line of data from the specified address to a location 2080309124Sdim/// closer to the processor. 2081309124Sdim/// 2082309124Sdim/// \headerfile <x86intrin.h> 2083309124Sdim/// 2084309124Sdim/// \code 2085309124Sdim/// void _mm_prefetch(const void * a, const int sel); 2086309124Sdim/// \endcode 2087309124Sdim/// 2088314564Sdim/// This intrinsic corresponds to the <c> PREFETCHNTA </c> instruction. 2089309124Sdim/// 2090309124Sdim/// \param a 2091309124Sdim/// A pointer to a memory location containing a cache line of data. 2092309124Sdim/// \param sel 2093314564Sdim/// A predefined integer constant specifying the type of prefetch 2094314564Sdim/// operation: \n 2095314564Sdim/// _MM_HINT_NTA: Move data using the non-temporal access (NTA) hint. The 2096314564Sdim/// PREFETCHNTA instruction will be generated. \n 2097309124Sdim/// _MM_HINT_T0: Move data using the T0 hint. The PREFETCHT0 instruction will 2098314564Sdim/// be generated. \n 2099309124Sdim/// _MM_HINT_T1: Move data using the T1 hint. The PREFETCHT1 instruction will 2100314564Sdim/// be generated. \n 2101309124Sdim/// _MM_HINT_T2: Move data using the T2 hint. The PREFETCHT2 instruction will 2102321369Sdim/// be generated. 2103327952Sdim#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), \ 2104327952Sdim ((sel) >> 2) & 1, (sel) & 0x3)) 2105276479Sdim#endif 2106193326Sed 2107341825Sdim/// Stores a 64-bit integer in the specified aligned memory location. To 2108309124Sdim/// minimize caching, the data is flagged as non-temporal (unlikely to be 2109309124Sdim/// used again soon). 2110309124Sdim/// 2111309124Sdim/// \headerfile <x86intrin.h> 2112309124Sdim/// 2113314564Sdim/// This intrinsic corresponds to the <c> MOVNTQ </c> instruction. 2114309124Sdim/// 2115309124Sdim/// \param __p 2116309124Sdim/// A pointer to an aligned memory location used to store the register value. 2117309124Sdim/// \param __a 2118309124Sdim/// A 64-bit integer containing the value to be stored. 2119341825Sdimstatic __inline__ void __DEFAULT_FN_ATTRS_MMX 2120249423Sdim_mm_stream_pi(__m64 *__p, __m64 __a) 2121193326Sed{ 2122249423Sdim __builtin_ia32_movntq(__p, __a); 2123193326Sed} 2124193326Sed 2125341825Sdim/// Moves packed float values from a 128-bit vector of [4 x float] to a 2126309124Sdim/// 128-bit aligned memory location. To minimize caching, the data is flagged 2127309124Sdim/// as non-temporal (unlikely to be used again soon). 2128309124Sdim/// 2129309124Sdim/// \headerfile <x86intrin.h> 2130309124Sdim/// 2131314564Sdim/// This intrinsic corresponds to the <c> VMOVNTPS / MOVNTPS </c> instruction. 2132309124Sdim/// 2133309124Sdim/// \param __p 2134309124Sdim/// A pointer to a 128-bit aligned memory location that will receive the 2135321369Sdim/// single-precision floating-point values. 2136309124Sdim/// \param __a 2137309124Sdim/// A 128-bit vector of [4 x float] containing the values to be moved. 2138288943Sdimstatic __inline__ void __DEFAULT_FN_ATTRS 2139249423Sdim_mm_stream_ps(float *__p, __m128 __a) 2140193326Sed{ 2141309124Sdim __builtin_nontemporal_store((__v4sf)__a, (__v4sf*)__p); 2142193326Sed} 2143193326Sed 2144314564Sdim#if defined(__cplusplus) 2145314564Sdimextern "C" { 2146314564Sdim#endif 2147314564Sdim 2148341825Sdim/// Forces strong memory ordering (serialization) between store 2149309124Sdim/// instructions preceding this instruction and store instructions following 2150309124Sdim/// this instruction, ensuring the system completes all previous stores 2151309124Sdim/// before executing subsequent stores. 2152309124Sdim/// 2153309124Sdim/// \headerfile <x86intrin.h> 2154309124Sdim/// 2155314564Sdim/// This intrinsic corresponds to the <c> SFENCE </c> instruction. 2156309124Sdim/// 2157314564Sdimvoid _mm_sfence(void); 2158193326Sed 2159314564Sdim#if defined(__cplusplus) 2160314564Sdim} // extern "C" 2161314564Sdim#endif 2162314564Sdim 2163341825Sdim/// Extracts 16-bit element from a 64-bit vector of [4 x i16] and 2164309124Sdim/// returns it, as specified by the immediate integer operand. 2165309124Sdim/// 2166309124Sdim/// \headerfile <x86intrin.h> 2167309124Sdim/// 2168314564Sdim/// \code 2169321369Sdim/// int _mm_extract_pi16(__m64 a, int n); 2170314564Sdim/// \endcode 2171309124Sdim/// 2172314564Sdim/// This intrinsic corresponds to the <c> VPEXTRW / PEXTRW </c> instruction. 2173314564Sdim/// 2174314564Sdim/// \param a 2175309124Sdim/// A 64-bit vector of [4 x i16]. 2176314564Sdim/// \param n 2177314564Sdim/// An immediate integer operand that determines which bits are extracted: \n 2178314564Sdim/// 0: Bits [15:0] are copied to the destination. \n 2179314564Sdim/// 1: Bits [31:16] are copied to the destination. \n 2180314564Sdim/// 2: Bits [47:32] are copied to the destination. \n 2181309124Sdim/// 3: Bits [63:48] are copied to the destination. 2182309124Sdim/// \returns A 16-bit integer containing the extracted 16 bits of packed data. 2183341825Sdim#define _mm_extract_pi16(a, n) \ 2184341825Sdim (int)__builtin_ia32_vec_ext_v4hi((__m64)a, (int)n) 2185193326Sed 2186341825Sdim/// Copies data from the 64-bit vector of [4 x i16] to the destination, 2187309124Sdim/// and inserts the lower 16-bits of an integer operand at the 16-bit offset 2188314564Sdim/// specified by the immediate operand \a n. 2189309124Sdim/// 2190309124Sdim/// \headerfile <x86intrin.h> 2191309124Sdim/// 2192314564Sdim/// \code 2193321369Sdim/// __m64 _mm_insert_pi16(__m64 a, int d, int n); 2194314564Sdim/// \endcode 2195309124Sdim/// 2196341825Sdim/// This intrinsic corresponds to the <c> PINSRW </c> instruction. 2197314564Sdim/// 2198314564Sdim/// \param a 2199309124Sdim/// A 64-bit vector of [4 x i16]. 2200314564Sdim/// \param d 2201309124Sdim/// An integer. The lower 16-bit value from this operand is written to the 2202314564Sdim/// destination at the offset specified by operand \a n. 2203314564Sdim/// \param n 2204309124Sdim/// An immediate integer operant that determines which the bits to be used 2205314564Sdim/// in the destination. \n 2206314564Sdim/// 0: Bits [15:0] are copied to the destination. \n 2207314564Sdim/// 1: Bits [31:16] are copied to the destination. \n 2208314564Sdim/// 2: Bits [47:32] are copied to the destination. \n 2209314564Sdim/// 3: Bits [63:48] are copied to the destination. \n 2210309124Sdim/// The remaining bits in the destination are copied from the corresponding 2211314564Sdim/// bits in operand \a a. 2212309124Sdim/// \returns A 64-bit integer vector containing the copied packed data from the 2213309124Sdim/// operands. 2214341825Sdim#define _mm_insert_pi16(a, d, n) \ 2215341825Sdim (__m64)__builtin_ia32_vec_set_v4hi((__m64)a, (int)d, (int)n) 2216193326Sed 2217341825Sdim/// Compares each of the corresponding packed 16-bit integer values of 2218309124Sdim/// the 64-bit integer vectors, and writes the greater value to the 2219309124Sdim/// corresponding bits in the destination. 2220309124Sdim/// 2221309124Sdim/// \headerfile <x86intrin.h> 2222309124Sdim/// 2223314564Sdim/// This intrinsic corresponds to the <c> PMAXSW </c> instruction. 2224309124Sdim/// 2225309124Sdim/// \param __a 2226309124Sdim/// A 64-bit integer vector containing one of the source operands. 2227309124Sdim/// \param __b 2228309124Sdim/// A 64-bit integer vector containing one of the source operands. 2229309124Sdim/// \returns A 64-bit integer vector containing the comparison results. 2230341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2231249423Sdim_mm_max_pi16(__m64 __a, __m64 __b) 2232193326Sed{ 2233249423Sdim return (__m64)__builtin_ia32_pmaxsw((__v4hi)__a, (__v4hi)__b); 2234193326Sed} 2235193326Sed 2236341825Sdim/// Compares each of the corresponding packed 8-bit unsigned integer 2237309124Sdim/// values of the 64-bit integer vectors, and writes the greater value to the 2238309124Sdim/// corresponding bits in the destination. 2239309124Sdim/// 2240309124Sdim/// \headerfile <x86intrin.h> 2241309124Sdim/// 2242314564Sdim/// This intrinsic corresponds to the <c> PMAXUB </c> instruction. 2243309124Sdim/// 2244309124Sdim/// \param __a 2245309124Sdim/// A 64-bit integer vector containing one of the source operands. 2246309124Sdim/// \param __b 2247309124Sdim/// A 64-bit integer vector containing one of the source operands. 2248309124Sdim/// \returns A 64-bit integer vector containing the comparison results. 2249341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2250249423Sdim_mm_max_pu8(__m64 __a, __m64 __b) 2251193326Sed{ 2252249423Sdim return (__m64)__builtin_ia32_pmaxub((__v8qi)__a, (__v8qi)__b); 2253193326Sed} 2254193326Sed 2255341825Sdim/// Compares each of the corresponding packed 16-bit integer values of 2256309124Sdim/// the 64-bit integer vectors, and writes the lesser value to the 2257309124Sdim/// corresponding bits in the destination. 2258309124Sdim/// 2259309124Sdim/// \headerfile <x86intrin.h> 2260309124Sdim/// 2261314564Sdim/// This intrinsic corresponds to the <c> PMINSW </c> instruction. 2262309124Sdim/// 2263309124Sdim/// \param __a 2264309124Sdim/// A 64-bit integer vector containing one of the source operands. 2265309124Sdim/// \param __b 2266309124Sdim/// A 64-bit integer vector containing one of the source operands. 2267309124Sdim/// \returns A 64-bit integer vector containing the comparison results. 2268341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2269249423Sdim_mm_min_pi16(__m64 __a, __m64 __b) 2270193326Sed{ 2271249423Sdim return (__m64)__builtin_ia32_pminsw((__v4hi)__a, (__v4hi)__b); 2272193326Sed} 2273193326Sed 2274341825Sdim/// Compares each of the corresponding packed 8-bit unsigned integer 2275309124Sdim/// values of the 64-bit integer vectors, and writes the lesser value to the 2276309124Sdim/// corresponding bits in the destination. 2277309124Sdim/// 2278309124Sdim/// \headerfile <x86intrin.h> 2279309124Sdim/// 2280314564Sdim/// This intrinsic corresponds to the <c> PMINUB </c> instruction. 2281309124Sdim/// 2282309124Sdim/// \param __a 2283309124Sdim/// A 64-bit integer vector containing one of the source operands. 2284309124Sdim/// \param __b 2285309124Sdim/// A 64-bit integer vector containing one of the source operands. 2286309124Sdim/// \returns A 64-bit integer vector containing the comparison results. 2287341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2288249423Sdim_mm_min_pu8(__m64 __a, __m64 __b) 2289193326Sed{ 2290249423Sdim return (__m64)__builtin_ia32_pminub((__v8qi)__a, (__v8qi)__b); 2291193326Sed} 2292193326Sed 2293341825Sdim/// Takes the most significant bit from each 8-bit element in a 64-bit 2294341825Sdim/// integer vector to create an 8-bit mask value. Zero-extends the value to 2295309124Sdim/// 32-bit integer and writes it to the destination. 2296309124Sdim/// 2297309124Sdim/// \headerfile <x86intrin.h> 2298309124Sdim/// 2299314564Sdim/// This intrinsic corresponds to the <c> PMOVMSKB </c> instruction. 2300309124Sdim/// 2301309124Sdim/// \param __a 2302309124Sdim/// A 64-bit integer vector containing the values with bits to be extracted. 2303341825Sdim/// \returns The most significant bit from each 8-bit element in \a __a, 2304341825Sdim/// written to bits [7:0]. 2305341825Sdimstatic __inline__ int __DEFAULT_FN_ATTRS_MMX 2306249423Sdim_mm_movemask_pi8(__m64 __a) 2307193326Sed{ 2308249423Sdim return __builtin_ia32_pmovmskb((__v8qi)__a); 2309193326Sed} 2310193326Sed 2311341825Sdim/// Multiplies packed 16-bit unsigned integer values and writes the 2312309124Sdim/// high-order 16 bits of each 32-bit product to the corresponding bits in 2313309124Sdim/// the destination. 2314309124Sdim/// 2315309124Sdim/// \headerfile <x86intrin.h> 2316309124Sdim/// 2317314564Sdim/// This intrinsic corresponds to the <c> PMULHUW </c> instruction. 2318309124Sdim/// 2319309124Sdim/// \param __a 2320309124Sdim/// A 64-bit integer vector containing one of the source operands. 2321309124Sdim/// \param __b 2322309124Sdim/// A 64-bit integer vector containing one of the source operands. 2323309124Sdim/// \returns A 64-bit integer vector containing the products of both operands. 2324341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2325249423Sdim_mm_mulhi_pu16(__m64 __a, __m64 __b) 2326193326Sed{ 2327249423Sdim return (__m64)__builtin_ia32_pmulhuw((__v4hi)__a, (__v4hi)__b); 2328193326Sed} 2329193326Sed 2330341825Sdim/// Shuffles the 4 16-bit integers from a 64-bit integer vector to the 2331309124Sdim/// destination, as specified by the immediate value operand. 2332309124Sdim/// 2333309124Sdim/// \headerfile <x86intrin.h> 2334309124Sdim/// 2335309124Sdim/// \code 2336309124Sdim/// __m64 _mm_shuffle_pi16(__m64 a, const int n); 2337309124Sdim/// \endcode 2338309124Sdim/// 2339314564Sdim/// This intrinsic corresponds to the <c> PSHUFW </c> instruction. 2340314564Sdim/// 2341309124Sdim/// \param a 2342309124Sdim/// A 64-bit integer vector containing the values to be shuffled. 2343309124Sdim/// \param n 2344309124Sdim/// An immediate value containing an 8-bit value specifying which elements to 2345314564Sdim/// copy from \a a. The destinations within the 64-bit destination are 2346314564Sdim/// assigned values as follows: \n 2347314564Sdim/// Bits [1:0] are used to assign values to bits [15:0] in the 2348314564Sdim/// destination. \n 2349314564Sdim/// Bits [3:2] are used to assign values to bits [31:16] in the 2350314564Sdim/// destination. \n 2351314564Sdim/// Bits [5:4] are used to assign values to bits [47:32] in the 2352314564Sdim/// destination. \n 2353314564Sdim/// Bits [7:6] are used to assign values to bits [63:48] in the 2354314564Sdim/// destination. \n 2355314564Sdim/// Bit value assignments: \n 2356314564Sdim/// 00: assigned from bits [15:0] of \a a. \n 2357314564Sdim/// 01: assigned from bits [31:16] of \a a. \n 2358314564Sdim/// 10: assigned from bits [47:32] of \a a. \n 2359314564Sdim/// 11: assigned from bits [63:48] of \a a. 2360309124Sdim/// \returns A 64-bit integer vector containing the shuffled values. 2361341825Sdim#define _mm_shuffle_pi16(a, n) \ 2362341825Sdim (__m64)__builtin_ia32_pshufw((__v4hi)(__m64)(a), (n)) 2363193326Sed 2364341825Sdim/// Conditionally copies the values from each 8-bit element in the first 2365309124Sdim/// 64-bit integer vector operand to the specified memory location, as 2366309124Sdim/// specified by the most significant bit in the corresponding element in the 2367321369Sdim/// second 64-bit integer vector operand. 2368309124Sdim/// 2369321369Sdim/// To minimize caching, the data is flagged as non-temporal 2370321369Sdim/// (unlikely to be used again soon). 2371321369Sdim/// 2372309124Sdim/// \headerfile <x86intrin.h> 2373309124Sdim/// 2374314564Sdim/// This intrinsic corresponds to the <c> MASKMOVQ </c> instruction. 2375309124Sdim/// 2376309124Sdim/// \param __d 2377309124Sdim/// A 64-bit integer vector containing the values with elements to be copied. 2378309124Sdim/// \param __n 2379309124Sdim/// A 64-bit integer vector operand. The most significant bit from each 8-bit 2380314564Sdim/// element determines whether the corresponding element in operand \a __d 2381314564Sdim/// is copied. If the most significant bit of a given element is 1, the 2382314564Sdim/// corresponding element in operand \a __d is copied. 2383309124Sdim/// \param __p 2384309124Sdim/// A pointer to a 64-bit memory location that will receive the conditionally 2385309124Sdim/// copied integer values. The address of the memory location does not have 2386309124Sdim/// to be aligned. 2387341825Sdimstatic __inline__ void __DEFAULT_FN_ATTRS_MMX 2388249423Sdim_mm_maskmove_si64(__m64 __d, __m64 __n, char *__p) 2389193326Sed{ 2390249423Sdim __builtin_ia32_maskmovq((__v8qi)__d, (__v8qi)__n, __p); 2391193326Sed} 2392193326Sed 2393341825Sdim/// Computes the rounded averages of the packed unsigned 8-bit integer 2394309124Sdim/// values and writes the averages to the corresponding bits in the 2395309124Sdim/// destination. 2396309124Sdim/// 2397309124Sdim/// \headerfile <x86intrin.h> 2398309124Sdim/// 2399314564Sdim/// This intrinsic corresponds to the <c> PAVGB </c> instruction. 2400309124Sdim/// 2401309124Sdim/// \param __a 2402309124Sdim/// A 64-bit integer vector containing one of the source operands. 2403309124Sdim/// \param __b 2404309124Sdim/// A 64-bit integer vector containing one of the source operands. 2405309124Sdim/// \returns A 64-bit integer vector containing the averages of both operands. 2406341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2407249423Sdim_mm_avg_pu8(__m64 __a, __m64 __b) 2408193326Sed{ 2409249423Sdim return (__m64)__builtin_ia32_pavgb((__v8qi)__a, (__v8qi)__b); 2410193326Sed} 2411193326Sed 2412341825Sdim/// Computes the rounded averages of the packed unsigned 16-bit integer 2413309124Sdim/// values and writes the averages to the corresponding bits in the 2414309124Sdim/// destination. 2415309124Sdim/// 2416309124Sdim/// \headerfile <x86intrin.h> 2417309124Sdim/// 2418314564Sdim/// This intrinsic corresponds to the <c> PAVGW </c> instruction. 2419309124Sdim/// 2420309124Sdim/// \param __a 2421309124Sdim/// A 64-bit integer vector containing one of the source operands. 2422309124Sdim/// \param __b 2423309124Sdim/// A 64-bit integer vector containing one of the source operands. 2424309124Sdim/// \returns A 64-bit integer vector containing the averages of both operands. 2425341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2426249423Sdim_mm_avg_pu16(__m64 __a, __m64 __b) 2427193326Sed{ 2428249423Sdim return (__m64)__builtin_ia32_pavgw((__v4hi)__a, (__v4hi)__b); 2429193326Sed} 2430193326Sed 2431341825Sdim/// Subtracts the corresponding 8-bit unsigned integer values of the two 2432309124Sdim/// 64-bit vector operands and computes the absolute value for each of the 2433309124Sdim/// difference. Then sum of the 8 absolute differences is written to the 2434309124Sdim/// bits [15:0] of the destination; the remaining bits [63:16] are cleared. 2435309124Sdim/// 2436309124Sdim/// \headerfile <x86intrin.h> 2437309124Sdim/// 2438314564Sdim/// This intrinsic corresponds to the <c> PSADBW </c> instruction. 2439309124Sdim/// 2440309124Sdim/// \param __a 2441309124Sdim/// A 64-bit integer vector containing one of the source operands. 2442309124Sdim/// \param __b 2443309124Sdim/// A 64-bit integer vector containing one of the source operands. 2444309124Sdim/// \returns A 64-bit integer vector whose lower 16 bits contain the sums of the 2445309124Sdim/// sets of absolute differences between both operands. The upper bits are 2446309124Sdim/// cleared. 2447341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2448249423Sdim_mm_sad_pu8(__m64 __a, __m64 __b) 2449193326Sed{ 2450249423Sdim return (__m64)__builtin_ia32_psadbw((__v8qi)__a, (__v8qi)__b); 2451193326Sed} 2452193326Sed 2453314564Sdim#if defined(__cplusplus) 2454314564Sdimextern "C" { 2455314564Sdim#endif 2456314564Sdim 2457341825Sdim/// Returns the contents of the MXCSR register as a 32-bit unsigned 2458314564Sdim/// integer value. 2459314564Sdim/// 2460314564Sdim/// There are several groups of macros associated with this 2461309124Sdim/// intrinsic, including: 2462314564Sdim/// <ul> 2463314564Sdim/// <li> 2464314564Sdim/// For checking exception states: _MM_EXCEPT_INVALID, _MM_EXCEPT_DIV_ZERO, 2465309124Sdim/// _MM_EXCEPT_DENORM, _MM_EXCEPT_OVERFLOW, _MM_EXCEPT_UNDERFLOW, 2466309124Sdim/// _MM_EXCEPT_INEXACT. There is a convenience wrapper 2467309124Sdim/// _MM_GET_EXCEPTION_STATE(). 2468314564Sdim/// </li> 2469314564Sdim/// <li> 2470314564Sdim/// For checking exception masks: _MM_MASK_UNDERFLOW, _MM_MASK_OVERFLOW, 2471309124Sdim/// _MM_MASK_INVALID, _MM_MASK_DENORM, _MM_MASK_DIV_ZERO, _MM_MASK_INEXACT. 2472309124Sdim/// There is a convenience wrapper _MM_GET_EXCEPTION_MASK(). 2473321369Sdim/// </li> 2474314564Sdim/// <li> 2475314564Sdim/// For checking rounding modes: _MM_ROUND_NEAREST, _MM_ROUND_DOWN, 2476309124Sdim/// _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO. There is a convenience wrapper 2477341825Sdim/// _MM_GET_ROUNDING_MODE(). 2478314564Sdim/// </li> 2479321369Sdim/// <li> 2480314564Sdim/// For checking flush-to-zero mode: _MM_FLUSH_ZERO_ON, _MM_FLUSH_ZERO_OFF. 2481309124Sdim/// There is a convenience wrapper _MM_GET_FLUSH_ZERO_MODE(). 2482314564Sdim/// </li> 2483321369Sdim/// <li> 2484314564Sdim/// For checking denormals-are-zero mode: _MM_DENORMALS_ZERO_ON, 2485309124Sdim/// _MM_DENORMALS_ZERO_OFF. There is a convenience wrapper 2486309124Sdim/// _MM_GET_DENORMALS_ZERO_MODE(). 2487314564Sdim/// </li> 2488314564Sdim/// </ul> 2489309124Sdim/// 2490341825Sdim/// For example, the following expression checks if an overflow exception has 2491309124Sdim/// occurred: 2492341825Sdim/// \code 2493309124Sdim/// ( _mm_getcsr() & _MM_EXCEPT_OVERFLOW ) 2494341825Sdim/// \endcode 2495309124Sdim/// 2496341825Sdim/// The following expression gets the current rounding mode: 2497341825Sdim/// \code 2498309124Sdim/// _MM_GET_ROUNDING_MODE() 2499341825Sdim/// \endcode 2500309124Sdim/// 2501309124Sdim/// \headerfile <x86intrin.h> 2502309124Sdim/// 2503314564Sdim/// This intrinsic corresponds to the <c> VSTMXCSR / STMXCSR </c> instruction. 2504309124Sdim/// 2505309124Sdim/// \returns A 32-bit unsigned integer containing the contents of the MXCSR 2506309124Sdim/// register. 2507314564Sdimunsigned int _mm_getcsr(void); 2508193326Sed 2509341825Sdim/// Sets the MXCSR register with the 32-bit unsigned integer value. 2510321369Sdim/// 2511314564Sdim/// There are several groups of macros associated with this intrinsic, 2512314564Sdim/// including: 2513314564Sdim/// <ul> 2514321369Sdim/// <li> 2515314564Sdim/// For setting exception states: _MM_EXCEPT_INVALID, _MM_EXCEPT_DIV_ZERO, 2516309124Sdim/// _MM_EXCEPT_DENORM, _MM_EXCEPT_OVERFLOW, _MM_EXCEPT_UNDERFLOW, 2517309124Sdim/// _MM_EXCEPT_INEXACT. There is a convenience wrapper 2518309124Sdim/// _MM_SET_EXCEPTION_STATE(x) where x is one of these macros. 2519314564Sdim/// </li> 2520314564Sdim/// <li> 2521314564Sdim/// For setting exception masks: _MM_MASK_UNDERFLOW, _MM_MASK_OVERFLOW, 2522309124Sdim/// _MM_MASK_INVALID, _MM_MASK_DENORM, _MM_MASK_DIV_ZERO, _MM_MASK_INEXACT. 2523309124Sdim/// There is a convenience wrapper _MM_SET_EXCEPTION_MASK(x) where x is one 2524309124Sdim/// of these macros. 2525314564Sdim/// </li> 2526314564Sdim/// <li> 2527314564Sdim/// For setting rounding modes: _MM_ROUND_NEAREST, _MM_ROUND_DOWN, 2528309124Sdim/// _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO. There is a convenience wrapper 2529309124Sdim/// _MM_SET_ROUNDING_MODE(x) where x is one of these macros. 2530314564Sdim/// </li> 2531314564Sdim/// <li> 2532314564Sdim/// For setting flush-to-zero mode: _MM_FLUSH_ZERO_ON, _MM_FLUSH_ZERO_OFF. 2533309124Sdim/// There is a convenience wrapper _MM_SET_FLUSH_ZERO_MODE(x) where x is 2534309124Sdim/// one of these macros. 2535314564Sdim/// </li> 2536314564Sdim/// <li> 2537314564Sdim/// For setting denormals-are-zero mode: _MM_DENORMALS_ZERO_ON, 2538309124Sdim/// _MM_DENORMALS_ZERO_OFF. There is a convenience wrapper 2539309124Sdim/// _MM_SET_DENORMALS_ZERO_MODE(x) where x is one of these macros. 2540314564Sdim/// </li> 2541314564Sdim/// </ul> 2542309124Sdim/// 2543309124Sdim/// For example, the following expression causes subsequent floating-point 2544309124Sdim/// operations to round up: 2545309124Sdim/// _mm_setcsr(_mm_getcsr() | _MM_ROUND_UP) 2546309124Sdim/// 2547309124Sdim/// The following example sets the DAZ and FTZ flags: 2548341825Sdim/// \code 2549341825Sdim/// void setFlags() { 2550341825Sdim/// _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON); 2551341825Sdim/// _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON); 2552341825Sdim/// } 2553341825Sdim/// \endcode 2554309124Sdim/// 2555309124Sdim/// \headerfile <x86intrin.h> 2556309124Sdim/// 2557314564Sdim/// This intrinsic corresponds to the <c> VLDMXCSR / LDMXCSR </c> instruction. 2558309124Sdim/// 2559309124Sdim/// \param __i 2560309124Sdim/// A 32-bit unsigned integer value to be written to the MXCSR register. 2561321369Sdimvoid _mm_setcsr(unsigned int __i); 2562193326Sed 2563314564Sdim#if defined(__cplusplus) 2564314564Sdim} // extern "C" 2565314564Sdim#endif 2566314564Sdim 2567341825Sdim/// Selects 4 float values from the 128-bit operands of [4 x float], as 2568309124Sdim/// specified by the immediate value operand. 2569309124Sdim/// 2570309124Sdim/// \headerfile <x86intrin.h> 2571309124Sdim/// 2572309124Sdim/// \code 2573309124Sdim/// __m128 _mm_shuffle_ps(__m128 a, __m128 b, const int mask); 2574309124Sdim/// \endcode 2575309124Sdim/// 2576314564Sdim/// This intrinsic corresponds to the <c> VSHUFPS / SHUFPS </c> instruction. 2577309124Sdim/// 2578309124Sdim/// \param a 2579309124Sdim/// A 128-bit vector of [4 x float]. 2580309124Sdim/// \param b 2581309124Sdim/// A 128-bit vector of [4 x float]. 2582309124Sdim/// \param mask 2583309124Sdim/// An immediate value containing an 8-bit value specifying which elements to 2584321369Sdim/// copy from \a a and \a b. \n 2585314564Sdim/// Bits [3:0] specify the values copied from operand \a a. \n 2586314564Sdim/// Bits [7:4] specify the values copied from operand \a b. \n 2587314564Sdim/// The destinations within the 128-bit destination are assigned values as 2588314564Sdim/// follows: \n 2589314564Sdim/// Bits [1:0] are used to assign values to bits [31:0] in the 2590314564Sdim/// destination. \n 2591314564Sdim/// Bits [3:2] are used to assign values to bits [63:32] in the 2592314564Sdim/// destination. \n 2593314564Sdim/// Bits [5:4] are used to assign values to bits [95:64] in the 2594314564Sdim/// destination. \n 2595314564Sdim/// Bits [7:6] are used to assign values to bits [127:96] in the 2596314564Sdim/// destination. \n 2597314564Sdim/// Bit value assignments: \n 2598314564Sdim/// 00: Bits [31:0] copied from the specified operand. \n 2599314564Sdim/// 01: Bits [63:32] copied from the specified operand. \n 2600314564Sdim/// 10: Bits [95:64] copied from the specified operand. \n 2601309124Sdim/// 11: Bits [127:96] copied from the specified operand. 2602309124Sdim/// \returns A 128-bit vector of [4 x float] containing the shuffled values. 2603341825Sdim#define _mm_shuffle_ps(a, b, mask) \ 2604341825Sdim (__m128)__builtin_ia32_shufps((__v4sf)(__m128)(a), (__v4sf)(__m128)(b), \ 2605341825Sdim (int)(mask)) 2606193326Sed 2607341825Sdim/// Unpacks the high-order (index 2,3) values from two 128-bit vectors of 2608314564Sdim/// [4 x float] and interleaves them into a 128-bit vector of [4 x float]. 2609309124Sdim/// 2610309124Sdim/// \headerfile <x86intrin.h> 2611309124Sdim/// 2612314564Sdim/// This intrinsic corresponds to the <c> VUNPCKHPS / UNPCKHPS </c> instruction. 2613309124Sdim/// 2614309124Sdim/// \param __a 2615314564Sdim/// A 128-bit vector of [4 x float]. \n 2616314564Sdim/// Bits [95:64] are written to bits [31:0] of the destination. \n 2617309124Sdim/// Bits [127:96] are written to bits [95:64] of the destination. 2618309124Sdim/// \param __b 2619309124Sdim/// A 128-bit vector of [4 x float]. 2620314564Sdim/// Bits [95:64] are written to bits [63:32] of the destination. \n 2621309124Sdim/// Bits [127:96] are written to bits [127:96] of the destination. 2622309124Sdim/// \returns A 128-bit vector of [4 x float] containing the interleaved values. 2623288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 2624249423Sdim_mm_unpackhi_ps(__m128 __a, __m128 __b) 2625193326Sed{ 2626309124Sdim return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 2, 6, 3, 7); 2627193326Sed} 2628193326Sed 2629341825Sdim/// Unpacks the low-order (index 0,1) values from two 128-bit vectors of 2630314564Sdim/// [4 x float] and interleaves them into a 128-bit vector of [4 x float]. 2631309124Sdim/// 2632309124Sdim/// \headerfile <x86intrin.h> 2633309124Sdim/// 2634314564Sdim/// This intrinsic corresponds to the <c> VUNPCKLPS / UNPCKLPS </c> instruction. 2635309124Sdim/// 2636309124Sdim/// \param __a 2637314564Sdim/// A 128-bit vector of [4 x float]. \n 2638314564Sdim/// Bits [31:0] are written to bits [31:0] of the destination. \n 2639309124Sdim/// Bits [63:32] are written to bits [95:64] of the destination. 2640309124Sdim/// \param __b 2641314564Sdim/// A 128-bit vector of [4 x float]. \n 2642314564Sdim/// Bits [31:0] are written to bits [63:32] of the destination. \n 2643309124Sdim/// Bits [63:32] are written to bits [127:96] of the destination. 2644309124Sdim/// \returns A 128-bit vector of [4 x float] containing the interleaved values. 2645288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 2646249423Sdim_mm_unpacklo_ps(__m128 __a, __m128 __b) 2647193326Sed{ 2648309124Sdim return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 0, 4, 1, 5); 2649193326Sed} 2650193326Sed 2651341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float]. The lower 2652309124Sdim/// 32 bits are set to the lower 32 bits of the second parameter. The upper 2653309124Sdim/// 96 bits are set to the upper 96 bits of the first parameter. 2654309124Sdim/// 2655309124Sdim/// \headerfile <x86intrin.h> 2656309124Sdim/// 2657341825Sdim/// This intrinsic corresponds to the <c> VBLENDPS / BLENDPS / MOVSS </c> 2658341825Sdim/// instruction. 2659309124Sdim/// 2660309124Sdim/// \param __a 2661309124Sdim/// A 128-bit floating-point vector of [4 x float]. The upper 96 bits are 2662309124Sdim/// written to the upper 96 bits of the result. 2663309124Sdim/// \param __b 2664309124Sdim/// A 128-bit floating-point vector of [4 x float]. The lower 32 bits are 2665309124Sdim/// written to the lower 32 bits of the result. 2666309124Sdim/// \returns A 128-bit floating-point vector of [4 x float]. 2667288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 2668249423Sdim_mm_move_ss(__m128 __a, __m128 __b) 2669193326Sed{ 2670341825Sdim __a[0] = __b[0]; 2671341825Sdim return __a; 2672193326Sed} 2673193326Sed 2674341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float]. The lower 2675309124Sdim/// 64 bits are set to the upper 64 bits of the second parameter. The upper 2676309124Sdim/// 64 bits are set to the upper 64 bits of the first parameter. 2677309124Sdim/// 2678309124Sdim/// \headerfile <x86intrin.h> 2679309124Sdim/// 2680314564Sdim/// This intrinsic corresponds to the <c> VUNPCKHPD / UNPCKHPD </c> instruction. 2681309124Sdim/// 2682309124Sdim/// \param __a 2683309124Sdim/// A 128-bit floating-point vector of [4 x float]. The upper 64 bits are 2684309124Sdim/// written to the upper 64 bits of the result. 2685309124Sdim/// \param __b 2686309124Sdim/// A 128-bit floating-point vector of [4 x float]. The upper 64 bits are 2687309124Sdim/// written to the lower 64 bits of the result. 2688309124Sdim/// \returns A 128-bit floating-point vector of [4 x float]. 2689288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 2690249423Sdim_mm_movehl_ps(__m128 __a, __m128 __b) 2691193326Sed{ 2692309124Sdim return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 6, 7, 2, 3); 2693193326Sed} 2694193326Sed 2695341825Sdim/// Constructs a 128-bit floating-point vector of [4 x float]. The lower 2696309124Sdim/// 64 bits are set to the lower 64 bits of the first parameter. The upper 2697309124Sdim/// 64 bits are set to the lower 64 bits of the second parameter. 2698309124Sdim/// 2699309124Sdim/// \headerfile <x86intrin.h> 2700309124Sdim/// 2701314564Sdim/// This intrinsic corresponds to the <c> VUNPCKLPD / UNPCKLPD </c> instruction. 2702309124Sdim/// 2703309124Sdim/// \param __a 2704309124Sdim/// A 128-bit floating-point vector of [4 x float]. The lower 64 bits are 2705309124Sdim/// written to the lower 64 bits of the result. 2706309124Sdim/// \param __b 2707309124Sdim/// A 128-bit floating-point vector of [4 x float]. The lower 64 bits are 2708309124Sdim/// written to the upper 64 bits of the result. 2709309124Sdim/// \returns A 128-bit floating-point vector of [4 x float]. 2710288943Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS 2711249423Sdim_mm_movelh_ps(__m128 __a, __m128 __b) 2712193326Sed{ 2713309124Sdim return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 0, 1, 4, 5); 2714193326Sed} 2715193326Sed 2716341825Sdim/// Converts a 64-bit vector of [4 x i16] into a 128-bit vector of [4 x 2717309124Sdim/// float]. 2718309124Sdim/// 2719309124Sdim/// \headerfile <x86intrin.h> 2720309124Sdim/// 2721321369Sdim/// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. 2722309124Sdim/// 2723309124Sdim/// \param __a 2724309124Sdim/// A 64-bit vector of [4 x i16]. The elements of the destination are copied 2725309124Sdim/// from the corresponding elements in this operand. 2726309124Sdim/// \returns A 128-bit vector of [4 x float] containing the copied and converted 2727309124Sdim/// values from the operand. 2728341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 2729249423Sdim_mm_cvtpi16_ps(__m64 __a) 2730193326Sed{ 2731249423Sdim __m64 __b, __c; 2732249423Sdim __m128 __r; 2733193326Sed 2734249423Sdim __b = _mm_setzero_si64(); 2735249423Sdim __b = _mm_cmpgt_pi16(__b, __a); 2736249423Sdim __c = _mm_unpackhi_pi16(__a, __b); 2737249423Sdim __r = _mm_setzero_ps(); 2738249423Sdim __r = _mm_cvtpi32_ps(__r, __c); 2739249423Sdim __r = _mm_movelh_ps(__r, __r); 2740249423Sdim __c = _mm_unpacklo_pi16(__a, __b); 2741249423Sdim __r = _mm_cvtpi32_ps(__r, __c); 2742193326Sed 2743249423Sdim return __r; 2744193326Sed} 2745193326Sed 2746341825Sdim/// Converts a 64-bit vector of 16-bit unsigned integer values into a 2747309124Sdim/// 128-bit vector of [4 x float]. 2748309124Sdim/// 2749309124Sdim/// \headerfile <x86intrin.h> 2750309124Sdim/// 2751321369Sdim/// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. 2752309124Sdim/// 2753309124Sdim/// \param __a 2754309124Sdim/// A 64-bit vector of 16-bit unsigned integer values. The elements of the 2755309124Sdim/// destination are copied from the corresponding elements in this operand. 2756309124Sdim/// \returns A 128-bit vector of [4 x float] containing the copied and converted 2757309124Sdim/// values from the operand. 2758341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 2759249423Sdim_mm_cvtpu16_ps(__m64 __a) 2760193326Sed{ 2761249423Sdim __m64 __b, __c; 2762249423Sdim __m128 __r; 2763193326Sed 2764249423Sdim __b = _mm_setzero_si64(); 2765249423Sdim __c = _mm_unpackhi_pi16(__a, __b); 2766249423Sdim __r = _mm_setzero_ps(); 2767249423Sdim __r = _mm_cvtpi32_ps(__r, __c); 2768249423Sdim __r = _mm_movelh_ps(__r, __r); 2769249423Sdim __c = _mm_unpacklo_pi16(__a, __b); 2770249423Sdim __r = _mm_cvtpi32_ps(__r, __c); 2771193326Sed 2772249423Sdim return __r; 2773193326Sed} 2774193326Sed 2775341825Sdim/// Converts the lower four 8-bit values from a 64-bit vector of [8 x i8] 2776309124Sdim/// into a 128-bit vector of [4 x float]. 2777309124Sdim/// 2778309124Sdim/// \headerfile <x86intrin.h> 2779309124Sdim/// 2780321369Sdim/// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. 2781309124Sdim/// 2782309124Sdim/// \param __a 2783309124Sdim/// A 64-bit vector of [8 x i8]. The elements of the destination are copied 2784309124Sdim/// from the corresponding lower 4 elements in this operand. 2785309124Sdim/// \returns A 128-bit vector of [4 x float] containing the copied and converted 2786309124Sdim/// values from the operand. 2787341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 2788249423Sdim_mm_cvtpi8_ps(__m64 __a) 2789193326Sed{ 2790249423Sdim __m64 __b; 2791296417Sdim 2792249423Sdim __b = _mm_setzero_si64(); 2793249423Sdim __b = _mm_cmpgt_pi8(__b, __a); 2794249423Sdim __b = _mm_unpacklo_pi8(__a, __b); 2795193326Sed 2796249423Sdim return _mm_cvtpi16_ps(__b); 2797193326Sed} 2798193326Sed 2799341825Sdim/// Converts the lower four unsigned 8-bit integer values from a 64-bit 2800309124Sdim/// vector of [8 x u8] into a 128-bit vector of [4 x float]. 2801309124Sdim/// 2802309124Sdim/// \headerfile <x86intrin.h> 2803309124Sdim/// 2804321369Sdim/// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. 2805309124Sdim/// 2806309124Sdim/// \param __a 2807309124Sdim/// A 64-bit vector of unsigned 8-bit integer values. The elements of the 2808309124Sdim/// destination are copied from the corresponding lower 4 elements in this 2809309124Sdim/// operand. 2810309124Sdim/// \returns A 128-bit vector of [4 x float] containing the copied and converted 2811309124Sdim/// values from the source operand. 2812341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 2813249423Sdim_mm_cvtpu8_ps(__m64 __a) 2814193326Sed{ 2815249423Sdim __m64 __b; 2816296417Sdim 2817249423Sdim __b = _mm_setzero_si64(); 2818249423Sdim __b = _mm_unpacklo_pi8(__a, __b); 2819193326Sed 2820249423Sdim return _mm_cvtpi16_ps(__b); 2821193326Sed} 2822193326Sed 2823341825Sdim/// Converts the two 32-bit signed integer values from each 64-bit vector 2824309124Sdim/// operand of [2 x i32] into a 128-bit vector of [4 x float]. 2825309124Sdim/// 2826309124Sdim/// \headerfile <x86intrin.h> 2827309124Sdim/// 2828321369Sdim/// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. 2829309124Sdim/// 2830309124Sdim/// \param __a 2831309124Sdim/// A 64-bit vector of [2 x i32]. The lower elements of the destination are 2832309124Sdim/// copied from the elements in this operand. 2833309124Sdim/// \param __b 2834309124Sdim/// A 64-bit vector of [2 x i32]. The upper elements of the destination are 2835309124Sdim/// copied from the elements in this operand. 2836309124Sdim/// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the 2837309124Sdim/// copied and converted values from the first operand. The upper 64 bits 2838309124Sdim/// contain the copied and converted values from the second operand. 2839341825Sdimstatic __inline__ __m128 __DEFAULT_FN_ATTRS_MMX 2840249423Sdim_mm_cvtpi32x2_ps(__m64 __a, __m64 __b) 2841193326Sed{ 2842249423Sdim __m128 __c; 2843296417Sdim 2844249423Sdim __c = _mm_setzero_ps(); 2845249423Sdim __c = _mm_cvtpi32_ps(__c, __b); 2846249423Sdim __c = _mm_movelh_ps(__c, __c); 2847193326Sed 2848249423Sdim return _mm_cvtpi32_ps(__c, __a); 2849193326Sed} 2850193326Sed 2851341825Sdim/// Converts each single-precision floating-point element of a 128-bit 2852309124Sdim/// floating-point vector of [4 x float] into a 16-bit signed integer, and 2853321369Sdim/// packs the results into a 64-bit integer vector of [4 x i16]. 2854309124Sdim/// 2855321369Sdim/// If the floating-point element is NaN or infinity, or if the 2856321369Sdim/// floating-point element is greater than 0x7FFFFFFF or less than -0x8000, 2857321369Sdim/// it is converted to 0x8000. Otherwise if the floating-point element is 2858321369Sdim/// greater than 0x7FFF, it is converted to 0x7FFF. 2859321369Sdim/// 2860309124Sdim/// \headerfile <x86intrin.h> 2861309124Sdim/// 2862321369Sdim/// This intrinsic corresponds to the <c> CVTPS2PI + COMPOSITE </c> instruction. 2863309124Sdim/// 2864309124Sdim/// \param __a 2865309124Sdim/// A 128-bit floating-point vector of [4 x float]. 2866309124Sdim/// \returns A 64-bit integer vector of [4 x i16] containing the converted 2867309124Sdim/// values. 2868341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2869249423Sdim_mm_cvtps_pi16(__m128 __a) 2870193326Sed{ 2871249423Sdim __m64 __b, __c; 2872296417Sdim 2873249423Sdim __b = _mm_cvtps_pi32(__a); 2874249423Sdim __a = _mm_movehl_ps(__a, __a); 2875249423Sdim __c = _mm_cvtps_pi32(__a); 2876296417Sdim 2877266674Sdim return _mm_packs_pi32(__b, __c); 2878193326Sed} 2879193326Sed 2880341825Sdim/// Converts each single-precision floating-point element of a 128-bit 2881309124Sdim/// floating-point vector of [4 x float] into an 8-bit signed integer, and 2882309124Sdim/// packs the results into the lower 32 bits of a 64-bit integer vector of 2883321369Sdim/// [8 x i8]. The upper 32 bits of the vector are set to 0. 2884309124Sdim/// 2885321369Sdim/// If the floating-point element is NaN or infinity, or if the 2886321369Sdim/// floating-point element is greater than 0x7FFFFFFF or less than -0x80, it 2887321369Sdim/// is converted to 0x80. Otherwise if the floating-point element is greater 2888321369Sdim/// than 0x7F, it is converted to 0x7F. 2889321369Sdim/// 2890309124Sdim/// \headerfile <x86intrin.h> 2891309124Sdim/// 2892321369Sdim/// This intrinsic corresponds to the <c> CVTPS2PI + COMPOSITE </c> instruction. 2893309124Sdim/// 2894309124Sdim/// \param __a 2895309124Sdim/// 128-bit floating-point vector of [4 x float]. 2896309124Sdim/// \returns A 64-bit integer vector of [8 x i8]. The lower 32 bits contain the 2897309124Sdim/// converted values and the uppper 32 bits are set to zero. 2898341825Sdimstatic __inline__ __m64 __DEFAULT_FN_ATTRS_MMX 2899249423Sdim_mm_cvtps_pi8(__m128 __a) 2900193326Sed{ 2901249423Sdim __m64 __b, __c; 2902296417Sdim 2903249423Sdim __b = _mm_cvtps_pi16(__a); 2904249423Sdim __c = _mm_setzero_si64(); 2905296417Sdim 2906249423Sdim return _mm_packs_pi16(__b, __c); 2907193326Sed} 2908193326Sed 2909341825Sdim/// Extracts the sign bits from each single-precision floating-point 2910309124Sdim/// element of a 128-bit floating-point vector of [4 x float] and returns the 2911309124Sdim/// sign bits in bits [0:3] of the result. Bits [31:4] of the result are set 2912309124Sdim/// to zero. 2913309124Sdim/// 2914309124Sdim/// \headerfile <x86intrin.h> 2915309124Sdim/// 2916314564Sdim/// This intrinsic corresponds to the <c> VMOVMSKPS / MOVMSKPS </c> instruction. 2917309124Sdim/// 2918309124Sdim/// \param __a 2919309124Sdim/// A 128-bit floating-point vector of [4 x float]. 2920309124Sdim/// \returns A 32-bit integer value. Bits [3:0] contain the sign bits from each 2921309124Sdim/// single-precision floating-point element of the parameter. Bits [31:4] are 2922309124Sdim/// set to zero. 2923288943Sdimstatic __inline__ int __DEFAULT_FN_ATTRS 2924249423Sdim_mm_movemask_ps(__m128 __a) 2925193326Sed{ 2926309124Sdim return __builtin_ia32_movmskps((__v4sf)__a); 2927193326Sed} 2928193326Sed 2929296417Sdim 2930309124Sdim#define _MM_ALIGN16 __attribute__((aligned(16))) 2931296417Sdim 2932193326Sed#define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) 2933193326Sed 2934193326Sed#define _MM_EXCEPT_INVALID (0x0001) 2935193326Sed#define _MM_EXCEPT_DENORM (0x0002) 2936193326Sed#define _MM_EXCEPT_DIV_ZERO (0x0004) 2937193326Sed#define _MM_EXCEPT_OVERFLOW (0x0008) 2938193326Sed#define _MM_EXCEPT_UNDERFLOW (0x0010) 2939193326Sed#define _MM_EXCEPT_INEXACT (0x0020) 2940193326Sed#define _MM_EXCEPT_MASK (0x003f) 2941193326Sed 2942193326Sed#define _MM_MASK_INVALID (0x0080) 2943193326Sed#define _MM_MASK_DENORM (0x0100) 2944193326Sed#define _MM_MASK_DIV_ZERO (0x0200) 2945193326Sed#define _MM_MASK_OVERFLOW (0x0400) 2946193326Sed#define _MM_MASK_UNDERFLOW (0x0800) 2947193326Sed#define _MM_MASK_INEXACT (0x1000) 2948193326Sed#define _MM_MASK_MASK (0x1f80) 2949193326Sed 2950193326Sed#define _MM_ROUND_NEAREST (0x0000) 2951193326Sed#define _MM_ROUND_DOWN (0x2000) 2952193326Sed#define _MM_ROUND_UP (0x4000) 2953193326Sed#define _MM_ROUND_TOWARD_ZERO (0x6000) 2954193326Sed#define _MM_ROUND_MASK (0x6000) 2955193326Sed 2956193326Sed#define _MM_FLUSH_ZERO_MASK (0x8000) 2957193326Sed#define _MM_FLUSH_ZERO_ON (0x8000) 2958234353Sdim#define _MM_FLUSH_ZERO_OFF (0x0000) 2959193326Sed 2960193326Sed#define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK) 2961193326Sed#define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK) 2962193326Sed#define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK) 2963193326Sed#define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK) 2964193326Sed 2965193326Sed#define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x))) 2966193326Sed#define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x))) 2967193326Sed#define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x))) 2968193326Sed#define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x))) 2969193326Sed 2970193326Sed#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \ 2971193326Seddo { \ 2972193326Sed __m128 tmp3, tmp2, tmp1, tmp0; \ 2973193326Sed tmp0 = _mm_unpacklo_ps((row0), (row1)); \ 2974193326Sed tmp2 = _mm_unpacklo_ps((row2), (row3)); \ 2975193326Sed tmp1 = _mm_unpackhi_ps((row0), (row1)); \ 2976193326Sed tmp3 = _mm_unpackhi_ps((row2), (row3)); \ 2977193326Sed (row0) = _mm_movelh_ps(tmp0, tmp2); \ 2978193326Sed (row1) = _mm_movehl_ps(tmp2, tmp0); \ 2979193326Sed (row2) = _mm_movelh_ps(tmp1, tmp3); \ 2980203955Srdivacky (row3) = _mm_movehl_ps(tmp3, tmp1); \ 2981193326Sed} while (0) 2982193326Sed 2983212904Sdim/* Aliases for compatibility. */ 2984212904Sdim#define _m_pextrw _mm_extract_pi16 2985212904Sdim#define _m_pinsrw _mm_insert_pi16 2986212904Sdim#define _m_pmaxsw _mm_max_pi16 2987212904Sdim#define _m_pmaxub _mm_max_pu8 2988212904Sdim#define _m_pminsw _mm_min_pi16 2989212904Sdim#define _m_pminub _mm_min_pu8 2990212904Sdim#define _m_pmovmskb _mm_movemask_pi8 2991212904Sdim#define _m_pmulhuw _mm_mulhi_pu16 2992212904Sdim#define _m_pshufw _mm_shuffle_pi16 2993212904Sdim#define _m_maskmovq _mm_maskmove_si64 2994212904Sdim#define _m_pavgb _mm_avg_pu8 2995212904Sdim#define _m_pavgw _mm_avg_pu16 2996212904Sdim#define _m_psadbw _mm_sad_pu8 2997212904Sdim#define _m_ _mm_ 2998212904Sdim#define _m_ _mm_ 2999212904Sdim 3000288943Sdim#undef __DEFAULT_FN_ATTRS 3001341825Sdim#undef __DEFAULT_FN_ATTRS_MMX 3002288943Sdim 3003194179Sed/* Ugly hack for backwards-compatibility (compatible with gcc) */ 3004309124Sdim#if defined(__SSE2__) && !__building_module(_Builtin_intrinsics) 3005193326Sed#include <emmintrin.h> 3006194179Sed#endif 3007193326Sed 3008193326Sed#endif /* __XMMINTRIN_H */ 3009