1/* Copyright (C) 2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
2
3   This file is part of GCC.
4
5   GCC is free software; you can redistribute it and/or modify
6   it under the terms of the GNU General Public License as published by
7   the Free Software Foundation; either version 2, or (at your option)
8   any later version.
9
10   GCC is distributed in the hope that it will be useful,
11   but WITHOUT ANY WARRANTY; without even the implied warranty of
12   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13   GNU General Public License for more details.
14
15   You should have received a copy of the GNU General Public License
16   along with GCC; see the file COPYING.  If not, write to
17   the Free Software Foundation, 51 Franklin Street, Fifth Floor,
18   Boston, MA 02110-1301, USA.  */
19
20/* As a special exception, if you include this header file into source
21   files compiled by GCC, this header file does not by itself cause
22   the resulting executable to be covered by the GNU General Public
23   License.  This exception does not however invalidate any other
24   reasons why the executable file might be covered by the GNU General
25   Public License.  */
26
27/* Implemented from the specification included in the Intel C++ Compiler
28   User Guide and Reference, version 9.0.  */
29
30#ifndef _PMMINTRIN_H_INCLUDED
31#define _PMMINTRIN_H_INCLUDED
32
33#ifndef __SSE3__
34# error "SSE3 instruction set not enabled"
35#else
36
37/* We need definitions from the SSE2 and SSE header files*/
38#include <xmmintrin.h>
39#include <emmintrin.h>
40
41/* Additional bits in the MXCSR.  */
42#define _MM_DENORMALS_ZERO_MASK		0x0040
43#define _MM_DENORMALS_ZERO_ON		0x0040
44#define _MM_DENORMALS_ZERO_OFF		0x0000
45
46#define _MM_SET_DENORMALS_ZERO_MODE(mode) \
47  _mm_setcsr ((_mm_getcsr () & ~_MM_DENORMALS_ZERO_MASK) | (mode))
48#define _MM_GET_DENORMALS_ZERO_MODE() \
49  (_mm_getcsr() & _MM_DENORMALS_ZERO_MASK)
50
51static __inline __m128 __attribute__((__always_inline__))
52_mm_addsub_ps (__m128 __X, __m128 __Y)
53{
54  return (__m128) __builtin_ia32_addsubps ((__v4sf)__X, (__v4sf)__Y);
55}
56
57static __inline __m128 __attribute__((__always_inline__))
58_mm_hadd_ps (__m128 __X, __m128 __Y)
59{
60  return (__m128) __builtin_ia32_haddps ((__v4sf)__X, (__v4sf)__Y);
61}
62
63static __inline __m128 __attribute__((__always_inline__))
64_mm_hsub_ps (__m128 __X, __m128 __Y)
65{
66  return (__m128) __builtin_ia32_hsubps ((__v4sf)__X, (__v4sf)__Y);
67}
68
69static __inline __m128 __attribute__((__always_inline__))
70_mm_movehdup_ps (__m128 __X)
71{
72  return (__m128) __builtin_ia32_movshdup ((__v4sf)__X);
73}
74
75static __inline __m128 __attribute__((__always_inline__))
76_mm_moveldup_ps (__m128 __X)
77{
78  return (__m128) __builtin_ia32_movsldup ((__v4sf)__X);
79}
80
81static __inline __m128d __attribute__((__always_inline__))
82_mm_addsub_pd (__m128d __X, __m128d __Y)
83{
84  return (__m128d) __builtin_ia32_addsubpd ((__v2df)__X, (__v2df)__Y);
85}
86
87static __inline __m128d __attribute__((__always_inline__))
88_mm_hadd_pd (__m128d __X, __m128d __Y)
89{
90  return (__m128d) __builtin_ia32_haddpd ((__v2df)__X, (__v2df)__Y);
91}
92
93static __inline __m128d __attribute__((__always_inline__))
94_mm_hsub_pd (__m128d __X, __m128d __Y)
95{
96  return (__m128d) __builtin_ia32_hsubpd ((__v2df)__X, (__v2df)__Y);
97}
98
99static __inline __m128d __attribute__((__always_inline__))
100_mm_loaddup_pd (double const *__P)
101{
102  return _mm_load1_pd (__P);
103}
104
105static __inline __m128d __attribute__((__always_inline__))
106_mm_movedup_pd (__m128d __X)
107{
108  return _mm_shuffle_pd (__X, __X, _MM_SHUFFLE2 (0,0));
109}
110
111static __inline __m128i __attribute__((__always_inline__))
112_mm_lddqu_si128 (__m128i const *__P)
113{
114  return (__m128i) __builtin_ia32_lddqu ((char const *)__P);
115}
116
117static __inline void __attribute__((__always_inline__))
118_mm_monitor (void const * __P, unsigned int __E, unsigned int __H)
119{
120  __builtin_ia32_monitor (__P, __E, __H);
121}
122
123static __inline void __attribute__((__always_inline__))
124_mm_mwait (unsigned int __E, unsigned int __H)
125{
126  __builtin_ia32_mwait (__E, __H);
127}
128
129#endif /* __SSE3__ */
130
131#endif /* _PMMINTRIN_H_INCLUDED */
132