tree-optimization/114659 - VN and FP to int punning

[official-gcc.git] / gcc / config / i386 / xmmintrin.h

/* Copyright (C) 2002-2024 Free Software Foundation, Inc.

   This file is part of GCC.

   GCC is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3, or (at your option)
   any later version.

   GCC is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   Under Section 7 of GPL version 3, you are granted additional
   permissions described in the GCC Runtime Library Exception, version
   3.1, as published by the Free Software Foundation.

   You should have received a copy of the GNU General Public License and
   a copy of the GCC Runtime Library Exception along with this program;
   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
   <http://www.gnu.org/licenses/>.  */

/* Implemented from the specification included in the Intel C++ Compiler
   User Guide and Reference, version 9.0.  */

#ifndef _XMMINTRIN_H_INCLUDED
#define _XMMINTRIN_H_INCLUDED

/* We need type definitions from the MMX header file.  */
#include <mmintrin.h>

/* Get _mm_malloc () and _mm_free ().  */
#include <mm_malloc.h>

/* Constants for use with _mm_prefetch.  */
enum _mm_hint
{
  _MM_HINT_IT0 = 19,
  _MM_HINT_IT1 = 18,
  /* _MM_HINT_ET is _MM_HINT_T with set 3rd bit.  */
  _MM_HINT_ET0 = 7,
  _MM_HINT_T0 = 3,
  _MM_HINT_T1 = 2,
  _MM_HINT_T2 = 1,
  _MM_HINT_NTA = 0
};

/* Loads one cache line from address P to a location "closer" to the
   processor.  The selector I specifies the type of prefetch operation.  */
#ifdef __OPTIMIZE__
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_prefetch (const void *__P, enum _mm_hint __I)
{
  __builtin_ia32_prefetch (__P, (__I & 0x4) >> 2,
                           __I & 0x3, (__I & 0x10) >> 4);
}
#else
#define _mm_prefetch(P, I) \
  __builtin_ia32_prefetch ((P), ((I) & 0x4) >> 2, ((I) & 0x3), ((I) & 0x10) >> 4)
#endif

#ifndef __SSE__
#pragma GCC push_options
#pragma GCC target("sse")
#define __DISABLE_SSE__
#endif /* __SSE__ */

/* The Intel API is flexible enough that we must allow aliasing with other
   vector types, and their scalar components.  */
typedef float __m128 __attribute__ ((__vector_size__ (16), __may_alias__));

/* Unaligned version of the same type.  */
typedef float __m128_u __attribute__ ((__vector_size__ (16), __may_alias__, __aligned__ (1)));
typedef float __x86_float_u __attribute__ ((__may_alias__, __aligned__ (1)));

/* Internal data types for implementing the intrinsics.  */
typedef float __v4sf __attribute__ ((__vector_size__ (16)));

/* Create a selector for use with the SHUFPS instruction.  */
#define _MM_SHUFFLE(fp3,fp2,fp1,fp0) \
 (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))

/* Bits in the MXCSR.  */
#define _MM_EXCEPT_MASK       0x003f
#define _MM_EXCEPT_INVALID    0x0001
#define _MM_EXCEPT_DENORM     0x0002
#define _MM_EXCEPT_DIV_ZERO   0x0004
#define _MM_EXCEPT_OVERFLOW   0x0008
#define _MM_EXCEPT_UNDERFLOW  0x0010
#define _MM_EXCEPT_INEXACT    0x0020

#define _MM_MASK_MASK         0x1f80
#define _MM_MASK_INVALID      0x0080
#define _MM_MASK_DENORM       0x0100
#define _MM_MASK_DIV_ZERO     0x0200
#define _MM_MASK_OVERFLOW     0x0400
#define _MM_MASK_UNDERFLOW    0x0800
#define _MM_MASK_INEXACT      0x1000

#define _MM_ROUND_MASK        0x6000
#define _MM_ROUND_NEAREST     0x0000
#define _MM_ROUND_DOWN        0x2000
#define _MM_ROUND_UP          0x4000
#define _MM_ROUND_TOWARD_ZERO 0x6000

#define _MM_FLUSH_ZERO_MASK   0x8000
#define _MM_FLUSH_ZERO_ON     0x8000
#define _MM_FLUSH_ZERO_OFF    0x0000

/* Compare predicates for scalar and packed compare intrinsics.  */

/* Equal (ordered, non-signaling)  */
#define _CMP_EQ_OQ      0x00
/* Less-than (ordered, signaling)  */
#define _CMP_LT_OS      0x01
/* Less-than-or-equal (ordered, signaling)  */
#define _CMP_LE_OS      0x02
/* Unordered (non-signaling)  */
#define _CMP_UNORD_Q    0x03
/* Not-equal (unordered, non-signaling)  */
#define _CMP_NEQ_UQ     0x04
/* Not-less-than (unordered, signaling)  */
#define _CMP_NLT_US     0x05
/* Not-less-than-or-equal (unordered, signaling)  */
#define _CMP_NLE_US     0x06
/* Ordered (nonsignaling)   */
#define _CMP_ORD_Q      0x07

/* Create an undefined vector.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_undefined_ps (void)
{
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Winit-self"
  __m128 __Y = __Y;
#pragma GCC diagnostic pop
  return __Y;
}

/* Create a vector of zeros.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setzero_ps (void)
{
  return __extension__ (__m128){ 0.0f, 0.0f, 0.0f, 0.0f };
}

/* Perform the respective operation on the lower SPFP (single-precision
   floating-point) values of A and B; the upper three SPFP values are
   passed through from A.  */

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

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

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

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

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

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

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

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

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

/* Perform the respective operation on the four SPFP values in A and B.  */

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

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

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

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

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

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

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

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

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

/* Perform logical bit-wise operations on 128-bit values.  */

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

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

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

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

/* Perform a comparison on the lower SPFP values of A and B.  If the
   comparison is true, place a mask of all ones in the result, otherwise a
   mask of zeros.  The upper three SPFP values are passed through from A.  */

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

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

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

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

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

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

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

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

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

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

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

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

/* Perform a comparison on the four SPFP values of A and B.  For each
   element, if the comparison is true, place a mask of all ones in the
   result, otherwise a mask of zeros.  */

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

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

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

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

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

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

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

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

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

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

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

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

#ifdef __OPTIMIZE__
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmp_ps (__m128 __X, __m128 __Y, const int __P)
{
  return (__m128) __builtin_ia32_cmpps ((__v4sf)__X, (__v4sf)__Y, __P);
}

extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cmp_ss (__m128 __X, __m128 __Y, const int __P)
{
  return (__m128) __builtin_ia32_cmpss ((__v4sf)__X, (__v4sf)__Y, __P);
}
#else
#define _mm_cmp_ps(X, Y, P)                                             \
  ((__m128) __builtin_ia32_cmpps ((__v4sf)(__m128)(X),                  \
                                  (__v4sf)(__m128)(Y), (int)(P)))

#define _mm_cmp_ss(X, Y, P)                                             \
  ((__m128) __builtin_ia32_cmpss ((__v4sf)(__m128)(X),                  \
                                  (__v4sf)(__m128)(Y), (int)(P)))
#endif

/* Compare the lower SPFP values of A and B and return 1 if true
   and 0 if false.  */

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

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

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

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

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

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

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

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

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

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

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

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

/* Convert the lower SPFP value to a 32-bit integer according to the current
   rounding mode.  */
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si32 (__m128 __A)
{
  return __builtin_ia32_cvtss2si ((__v4sf) __A);
}

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

#ifdef __x86_64__
/* Convert the lower SPFP value to a 32-bit integer according to the
   current rounding mode.  */

/* Intel intrinsic.  */
extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si64 (__m128 __A)
{
  return __builtin_ia32_cvtss2si64 ((__v4sf) __A);
}

/* Microsoft intrinsic.  */
extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtss_si64x (__m128 __A)
{
  return __builtin_ia32_cvtss2si64 ((__v4sf) __A);
}
#endif

/* Convert the two lower SPFP values to 32-bit integers according to the
   current rounding mode.  Return the integers in packed form.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi32 (__m128 __A)
{
  return (__m64) __builtin_ia32_cvtps2pi ((__v4sf) __A);
}

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

/* Truncate the lower SPFP value to a 32-bit integer.  */
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si32 (__m128 __A)
{
  return __builtin_ia32_cvttss2si ((__v4sf) __A);
}

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

#ifdef __x86_64__
/* Truncate the lower SPFP value to a 32-bit integer.  */

/* Intel intrinsic.  */
extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si64 (__m128 __A)
{
  return __builtin_ia32_cvttss2si64 ((__v4sf) __A);
}

/* Microsoft intrinsic.  */
extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttss_si64x (__m128 __A)
{
  return __builtin_ia32_cvttss2si64 ((__v4sf) __A);
}
#endif

/* Truncate the two lower SPFP values to 32-bit integers.  Return the
   integers in packed form.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvttps_pi32 (__m128 __A)
{
  return (__m64) __builtin_ia32_cvttps2pi ((__v4sf) __A);
}

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

/* Convert B to a SPFP value and insert it as element zero in A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi32_ss (__m128 __A, int __B)
{
  return (__m128) __builtin_ia32_cvtsi2ss ((__v4sf) __A, __B);
}

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

#ifdef __x86_64__
/* Convert B to a SPFP value and insert it as element zero in A.  */

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

/* Microsoft intrinsic.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtsi64x_ss (__m128 __A, long long __B)
{
  return (__m128) __builtin_ia32_cvtsi642ss ((__v4sf) __A, __B);
}
#endif

/* Convert the two 32-bit values in B to SPFP form and insert them
   as the two lower elements in A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi32_ps (__m128 __A, __m64 __B)
{
  return (__m128) __builtin_ia32_cvtpi2ps ((__v4sf) __A, (__v2si)__B);
}

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

/* Convert the four signed 16-bit values in A to SPFP form.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi16_ps (__m64 __A)
{
  __v4hi __sign;
  __v2si __hisi, __losi;
  __v4sf __zero, __ra, __rb;

  /* This comparison against zero gives us a mask that can be used to
     fill in the missing sign bits in the unpack operations below, so
     that we get signed values after unpacking.  */
  __sign = __builtin_ia32_pcmpgtw ((__v4hi)0LL, (__v4hi)__A);

  /* Convert the four words to doublewords.  */
  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, __sign);
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, __sign);

  /* Convert the doublewords to floating point two at a time.  */
  __zero = (__v4sf) _mm_setzero_ps ();
  __ra = __builtin_ia32_cvtpi2ps (__zero, __losi);
  __rb = __builtin_ia32_cvtpi2ps (__ra, __hisi);

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

/* Convert the four unsigned 16-bit values in A to SPFP form.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpu16_ps (__m64 __A)
{
  __v2si __hisi, __losi;
  __v4sf __zero, __ra, __rb;

  /* Convert the four words to doublewords.  */
  __losi = (__v2si) __builtin_ia32_punpcklwd ((__v4hi)__A, (__v4hi)0LL);
  __hisi = (__v2si) __builtin_ia32_punpckhwd ((__v4hi)__A, (__v4hi)0LL);

  /* Convert the doublewords to floating point two at a time.  */
  __zero = (__v4sf) _mm_setzero_ps ();
  __ra = __builtin_ia32_cvtpi2ps (__zero, __losi);
  __rb = __builtin_ia32_cvtpi2ps (__ra, __hisi);

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

/* Convert the low four signed 8-bit values in A to SPFP form.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi8_ps (__m64 __A)
{
  __v8qi __sign;

  /* This comparison against zero gives us a mask that can be used to
     fill in the missing sign bits in the unpack operations below, so
     that we get signed values after unpacking.  */
  __sign = __builtin_ia32_pcmpgtb ((__v8qi)0LL, (__v8qi)__A);

  /* Convert the four low bytes to words.  */
  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, __sign);

  return _mm_cvtpi16_ps(__A);
}

/* Convert the low four unsigned 8-bit values in A to SPFP form.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpu8_ps(__m64 __A)
{
  __A = (__m64) __builtin_ia32_punpcklbw ((__v8qi)__A, (__v8qi)0LL);
  return _mm_cvtpu16_ps(__A);
}

/* Convert the four signed 32-bit values in A and B to SPFP form.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtpi32x2_ps(__m64 __A, __m64 __B)
{
  __v4sf __zero = (__v4sf) _mm_setzero_ps ();
  __v4sf __sfa = __builtin_ia32_cvtpi2ps (__zero, (__v2si)__A);
  __v4sf __sfb = __builtin_ia32_cvtpi2ps (__sfa, (__v2si)__B);
  return (__m128) __builtin_ia32_movlhps (__sfa, __sfb);
}

/* Convert the four SPFP values in A to four signed 16-bit integers.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi16(__m128 __A)
{
  __v4sf __hisf = (__v4sf)__A;
  __v4sf __losf = __builtin_ia32_movhlps (__hisf, __hisf);
  __v2si __hisi = __builtin_ia32_cvtps2pi (__hisf);
  __v2si __losi = __builtin_ia32_cvtps2pi (__losf);
  return (__m64) __builtin_ia32_packssdw (__hisi, __losi);
}

/* Convert the four SPFP values in A to four signed 8-bit integers.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_cvtps_pi8(__m128 __A)
{
  __v4hi __tmp = (__v4hi) _mm_cvtps_pi16 (__A);
  return (__m64) __builtin_ia32_packsswb (__tmp, (__v4hi)0LL);
}

/* Selects four specific SPFP values from A and B based on MASK.  */
#ifdef __OPTIMIZE__
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_ps (__m128 __A, __m128 __B, int const __mask)
{
  return (__m128) __builtin_ia32_shufps ((__v4sf)__A, (__v4sf)__B, __mask);
}
#else
#define _mm_shuffle_ps(A, B, MASK)                                      \
  ((__m128) __builtin_ia32_shufps ((__v4sf)(__m128)(A),                 \
                                   (__v4sf)(__m128)(B), (int)(MASK)))
#endif

/* Selects and interleaves the upper two SPFP values from A and B.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpackhi_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpckhps ((__v4sf)__A, (__v4sf)__B);
}

/* Selects and interleaves the lower two SPFP values from A and B.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_unpacklo_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_unpcklps ((__v4sf)__A, (__v4sf)__B);
}

/* Sets the upper two SPFP values with 64-bits of data loaded from P;
   the lower two values are passed through from A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadh_pi (__m128 __A, __m64_u const *__P)
{
  return (__m128) __builtin_ia32_loadhps ((__v4sf)__A, (const __v2sf *)__P);
}

/* Stores the upper two SPFP values of A into P.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeh_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storehps ((__v2sf *)__P, (__v4sf)__A);
}

/* Moves the upper two values of B into the lower two values of A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movehl_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movhlps ((__v4sf)__A, (__v4sf)__B);
}

/* Moves the lower two values of B into the upper two values of A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movelh_ps (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_ia32_movlhps ((__v4sf)__A, (__v4sf)__B);
}

/* Sets the lower two SPFP values with 64-bits of data loaded from P;
   the upper two values are passed through from A.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadl_pi (__m128 __A, __m64_u const *__P)
{
  return (__m128) __builtin_ia32_loadlps ((__v4sf)__A, (const __v2sf *)__P);
}

/* Stores the lower two SPFP values of A into P.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storel_pi (__m64 *__P, __m128 __A)
{
  __builtin_ia32_storelps ((__v2sf *)__P, (__v4sf)__A);
}

/* Creates a 4-bit mask from the most significant bits of the SPFP values.  */
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_ps (__m128 __A)
{
  return __builtin_ia32_movmskps ((__v4sf)__A);
}

/* Return the contents of the control register.  */
extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_getcsr (void)
{
  return __builtin_ia32_stmxcsr ();
}

/* Read exception bits from the control register.  */
extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_GET_EXCEPTION_STATE (void)
{
  return _mm_getcsr() & _MM_EXCEPT_MASK;
}

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

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

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

/* Set the control register to I.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setcsr (unsigned int __I)
{
  __builtin_ia32_ldmxcsr (__I);
}

/* Set exception bits in the control register.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_MM_SET_EXCEPTION_STATE(unsigned int __mask)
{
  _mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | __mask);
}

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

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

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

/* Create a vector with element 0 as F and the rest zero.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_ss (float __F)
{
  return __extension__ (__m128)(__v4sf){ __F, 0.0f, 0.0f, 0.0f };
}

/* Create a vector with all four elements equal to F.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set1_ps (float __F)
{
  return __extension__ (__m128)(__v4sf){ __F, __F, __F, __F };
}

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

/* Create a vector with element 0 as *P and the rest zero.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_ss (float const *__P)
{
  return __extension__ (__m128) (__v4sf){ *(__x86_float_u *)__P, 0.0f, 0.0f, 0.0f };
}

/* Create a vector with all four elements equal to *P.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load1_ps (float const *__P)
{
  return _mm_set1_ps (*__P);
}

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

/* Load four SPFP values from P.  The address must be 16-byte aligned.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_load_ps (float const *__P)
{
  return *(__m128 *)__P;
}

/* Load four SPFP values from P.  The address need not be 16-byte aligned.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadu_ps (float const *__P)
{
  return *(__m128_u *)__P;
}

/* Load four SPFP values in reverse order.  The address must be aligned.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_loadr_ps (float const *__P)
{
  __v4sf __tmp = *(__v4sf *)__P;
  return (__m128) __builtin_ia32_shufps (__tmp, __tmp, _MM_SHUFFLE (0,1,2,3));
}

/* Create the vector [Z Y X W].  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_set_ps (const float __Z, const float __Y, const float __X, const float __W)
{
  return __extension__ (__m128)(__v4sf){ __W, __X, __Y, __Z };
}

/* Create the vector [W X Y Z].  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_setr_ps (float __Z, float __Y, float __X, float __W)
{
  return __extension__ (__m128)(__v4sf){ __Z, __Y, __X, __W };
}

/* Stores the lower SPFP value.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_ss (float *__P, __m128 __A)
{
  *(__x86_float_u *)__P = ((__v4sf)__A)[0];
}

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

/* Store four SPFP values.  The address must be 16-byte aligned.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store_ps (float *__P, __m128 __A)
{
  *(__m128 *)__P = __A;
}

/* Store four SPFP values.  The address need not be 16-byte aligned.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storeu_ps (float *__P, __m128 __A)
{
  *(__m128_u *)__P = __A;
}

/* Store the lower SPFP value across four words.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_store1_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,0,0,0));
  _mm_storeu_ps (__P, __tmp);
}

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

/* Store four SPFP values in reverse order.  The address must be aligned.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_storer_ps (float *__P, __m128 __A)
{
  __v4sf __va = (__v4sf)__A;
  __v4sf __tmp = __builtin_ia32_shufps (__va, __va, _MM_SHUFFLE (0,1,2,3));
  _mm_store_ps (__P, __tmp);
}

/* Sets the low SPFP value of A from the low value of B.  */
extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_move_ss (__m128 __A, __m128 __B)
{
  return (__m128) __builtin_shuffle ((__v4sf)__A, (__v4sf)__B,
                                     __extension__
                                     (__attribute__((__vector_size__ (16))) int)
                                     {4,1,2,3});
}

/* Extracts one of the four words of A.  The selector N must be immediate.  */
#ifdef __OPTIMIZE__
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_extract_pi16 (__m64 const __A, int const __N)
{
  return (unsigned short) __builtin_ia32_vec_ext_v4hi ((__v4hi)__A, __N);
}

extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pextrw (__m64 const __A, int const __N)
{
  return _mm_extract_pi16 (__A, __N);
}
#else
#define _mm_extract_pi16(A, N)  \
  ((int) (unsigned short) __builtin_ia32_vec_ext_v4hi ((__v4hi)(__m64)(A), (int)(N)))

#define _m_pextrw(A, N) _mm_extract_pi16(A, N)
#endif

/* Inserts word D into one of four words of A.  The selector N must be
   immediate.  */
#ifdef __OPTIMIZE__
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_insert_pi16 (__m64 const __A, int const __D, int const __N)
{
  return (__m64) __builtin_ia32_vec_set_v4hi ((__v4hi)__A, __D, __N);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pinsrw (__m64 const __A, int const __D, int const __N)
{
  return _mm_insert_pi16 (__A, __D, __N);
}
#else
#define _mm_insert_pi16(A, D, N)                                \
  ((__m64) __builtin_ia32_vec_set_v4hi ((__v4hi)(__m64)(A),     \
                                        (int)(D), (int)(N)))

#define _m_pinsrw(A, D, N) _mm_insert_pi16(A, D, N)
#endif

/* Compute the element-wise maximum of signed 16-bit values.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxsw ((__v4hi)__A, (__v4hi)__B);
}

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

/* Compute the element-wise maximum of unsigned 8-bit values.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_max_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmaxub ((__v8qi)__A, (__v8qi)__B);
}

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

/* Compute the element-wise minimum of signed 16-bit values.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_pi16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminsw ((__v4hi)__A, (__v4hi)__B);
}

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

/* Compute the element-wise minimum of unsigned 8-bit values.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_min_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pminub ((__v8qi)__A, (__v8qi)__B);
}

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

/* Create an 8-bit mask of the signs of 8-bit values.  */
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_movemask_pi8 (__m64 __A)
{
  return __builtin_ia32_pmovmskb ((__v8qi)__A);
}

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

/* Multiply four unsigned 16-bit values in A by four unsigned 16-bit values
   in B and produce the high 16 bits of the 32-bit results.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_mulhi_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pmulhuw ((__v4hi)__A, (__v4hi)__B);
}

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

/* Return a combination of the four 16-bit values in A.  The selector
   must be an immediate.  */
#ifdef __OPTIMIZE__
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_shuffle_pi16 (__m64 __A, int const __N)
{
  return (__m64) __builtin_ia32_pshufw ((__v4hi)__A, __N);
}

extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_m_pshufw (__m64 __A, int const __N)
{
  return _mm_shuffle_pi16 (__A, __N);
}
#else
#define _mm_shuffle_pi16(A, N) \
  ((__m64) __builtin_ia32_pshufw ((__v4hi)(__m64)(A), (int)(N)))

#define _m_pshufw(A, N) _mm_shuffle_pi16 (A, N)
#endif

/* Conditionally store byte elements of A into P.  The high bit of each
   byte in the selector N determines whether the corresponding byte from
   A is stored.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_maskmove_si64 (__m64 __A, __m64 __N, char *__P)
{
#ifdef __MMX_WITH_SSE__
  /* Emulate MMX maskmovq with SSE2 maskmovdqu and handle unmapped bits
     64:127 at address __P.  */
  typedef long long __v2di __attribute__ ((__vector_size__ (16)));
  typedef char __v16qi __attribute__ ((__vector_size__ (16)));
  /* Zero-extend __A and __N to 128 bits.  */
  __v2di __A128 = __extension__ (__v2di) { ((__v1di) __A)[0], 0 };
  __v2di __N128 = __extension__ (__v2di) { ((__v1di) __N)[0], 0 };

  /* Check the alignment of __P.  */
  __SIZE_TYPE__ offset = ((__SIZE_TYPE__) __P) & 0xf;
  if (offset)
    {
      /* If the misalignment of __P > 8, subtract __P by 8 bytes.
         Otherwise, subtract __P by the misalignment.  */
      if (offset > 8)
        offset = 8;
      __P = (char *) (((__SIZE_TYPE__) __P) - offset);

      /* Shift __A128 and __N128 to the left by the adjustment.  */
      switch (offset)
        {
        case 1:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 8);
          break;
        case 2:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 2 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 2 * 8);
          break;
        case 3:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 3 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 3 * 8);
          break;
        case 4:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 4 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 4 * 8);
          break;
        case 5:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 5 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 5 * 8);
          break;
        case 6:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 6 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 6 * 8);
          break;
        case 7:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 7 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 7 * 8);
          break;
        case 8:
          __A128 = __builtin_ia32_pslldqi128 (__A128, 8 * 8);
          __N128 = __builtin_ia32_pslldqi128 (__N128, 8 * 8);
          break;
        default:
          break;
        }
    }
  __builtin_ia32_maskmovdqu ((__v16qi)__A128, (__v16qi)__N128, __P);
#else
  __builtin_ia32_maskmovq ((__v8qi)__A, (__v8qi)__N, __P);
#endif
}

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

/* Compute the rounded averages of the unsigned 8-bit values in A and B.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pavgb ((__v8qi)__A, (__v8qi)__B);
}

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

/* Compute the rounded averages of the unsigned 16-bit values in A and B.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_avg_pu16 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_pavgw ((__v4hi)__A, (__v4hi)__B);
}

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

/* Compute the sum of the absolute differences of the unsigned 8-bit
   values in A and B.  Return the value in the lower 16-bit word; the
   upper words are cleared.  */
extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sad_pu8 (__m64 __A, __m64 __B)
{
  return (__m64) __builtin_ia32_psadbw ((__v8qi)__A, (__v8qi)__B);
}

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

/* Stores the data in A to the address P without polluting the caches.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_pi (__m64 *__P, __m64 __A)
{
  __builtin_ia32_movntq ((unsigned long long *)__P, (unsigned long long)__A);
}

/* Likewise.  The address must be 16-byte aligned.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_stream_ps (float *__P, __m128 __A)
{
  __builtin_ia32_movntps (__P, (__v4sf)__A);
}

/* Guarantees that every preceding store is globally visible before
   any subsequent store.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sfence (void)
{
  __builtin_ia32_sfence ();
}

/* Transpose the 4x4 matrix composed of row[0-3].  */
#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3)                       \
do {                                                                    \
  __v4sf __r0 = (row0), __r1 = (row1), __r2 = (row2), __r3 = (row3);    \
  __v4sf __t0 = __builtin_ia32_unpcklps (__r0, __r1);                   \
  __v4sf __t1 = __builtin_ia32_unpcklps (__r2, __r3);                   \
  __v4sf __t2 = __builtin_ia32_unpckhps (__r0, __r1);                   \
  __v4sf __t3 = __builtin_ia32_unpckhps (__r2, __r3);                   \
  (row0) = __builtin_ia32_movlhps (__t0, __t1);                         \
  (row1) = __builtin_ia32_movhlps (__t1, __t0);                         \
  (row2) = __builtin_ia32_movlhps (__t2, __t3);                         \
  (row3) = __builtin_ia32_movhlps (__t3, __t2);                         \
} while (0)

/* For backward source compatibility.  */
# include <emmintrin.h>

#ifdef __DISABLE_SSE__
#undef __DISABLE_SSE__
#pragma GCC pop_options
#endif /* __DISABLE_SSE__ */

/* The execution of the next instruction is delayed by an implementation
   specific amount of time.  The instruction does not modify the
   architectural state.  This is after the pop_options pragma because
   it does not require SSE support in the processor--the encoding is a
   nop on processors that do not support it.  */
extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_pause (void)
{
  __builtin_ia32_pause ();
}

#endif /* _XMMINTRIN_H_INCLUDED */