1 /* Copyright (C) 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, 59 Temple Place - Suite 330,
18 Boston, MA 02111-1307, 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 9.1. */
29
30 #ifndef _TMMINTRIN_H_INCLUDED
31 #define _TMMINTRIN_H_INCLUDED
32
33 #ifndef __SSSE3__
34 # error "SSSE3 instruction set not enabled"
35 #else
36
37 /* We need definitions from the SSE3, SSE2 and SSE header files*/
38 #include <pmmintrin.h>
39
40 static __inline __m128i __attribute__((__always_inline__))
_mm_hadd_epi16(__m128i __X,__m128i __Y)41 _mm_hadd_epi16 (__m128i __X, __m128i __Y)
42 {
43 return (__m128i) __builtin_ia32_phaddw128 ((__v8hi)__X, (__v8hi)__Y);
44 }
45
46 static __inline __m128i __attribute__((__always_inline__))
_mm_hadd_epi32(__m128i __X,__m128i __Y)47 _mm_hadd_epi32 (__m128i __X, __m128i __Y)
48 {
49 return (__m128i) __builtin_ia32_phaddd128 ((__v4si)__X, (__v4si)__Y);
50 }
51
52 static __inline __m128i __attribute__((__always_inline__))
_mm_hadds_epi16(__m128i __X,__m128i __Y)53 _mm_hadds_epi16 (__m128i __X, __m128i __Y)
54 {
55 return (__m128i) __builtin_ia32_phaddsw128 ((__v8hi)__X, (__v8hi)__Y);
56 }
57
58 static __inline __m64 __attribute__((__always_inline__))
_mm_hadd_pi16(__m64 __X,__m64 __Y)59 _mm_hadd_pi16 (__m64 __X, __m64 __Y)
60 {
61 return (__m64) __builtin_ia32_phaddw ((__v4hi)__X, (__v4hi)__Y);
62 }
63
64 static __inline __m64 __attribute__((__always_inline__))
_mm_hadd_pi32(__m64 __X,__m64 __Y)65 _mm_hadd_pi32 (__m64 __X, __m64 __Y)
66 {
67 return (__m64) __builtin_ia32_phaddd ((__v2si)__X, (__v2si)__Y);
68 }
69
70 static __inline __m64 __attribute__((__always_inline__))
_mm_hadds_pi16(__m64 __X,__m64 __Y)71 _mm_hadds_pi16 (__m64 __X, __m64 __Y)
72 {
73 return (__m64) __builtin_ia32_phaddsw ((__v4hi)__X, (__v4hi)__Y);
74 }
75
76 static __inline __m128i __attribute__((__always_inline__))
_mm_hsub_epi16(__m128i __X,__m128i __Y)77 _mm_hsub_epi16 (__m128i __X, __m128i __Y)
78 {
79 return (__m128i) __builtin_ia32_phsubw128 ((__v8hi)__X, (__v8hi)__Y);
80 }
81
82 static __inline __m128i __attribute__((__always_inline__))
_mm_hsub_epi32(__m128i __X,__m128i __Y)83 _mm_hsub_epi32 (__m128i __X, __m128i __Y)
84 {
85 return (__m128i) __builtin_ia32_phsubd128 ((__v4si)__X, (__v4si)__Y);
86 }
87
88 static __inline __m128i __attribute__((__always_inline__))
_mm_hsubs_epi16(__m128i __X,__m128i __Y)89 _mm_hsubs_epi16 (__m128i __X, __m128i __Y)
90 {
91 return (__m128i) __builtin_ia32_phsubsw128 ((__v8hi)__X, (__v8hi)__Y);
92 }
93
94 static __inline __m64 __attribute__((__always_inline__))
_mm_hsub_pi16(__m64 __X,__m64 __Y)95 _mm_hsub_pi16 (__m64 __X, __m64 __Y)
96 {
97 return (__m64) __builtin_ia32_phsubw ((__v4hi)__X, (__v4hi)__Y);
98 }
99
100 static __inline __m64 __attribute__((__always_inline__))
_mm_hsub_pi32(__m64 __X,__m64 __Y)101 _mm_hsub_pi32 (__m64 __X, __m64 __Y)
102 {
103 return (__m64) __builtin_ia32_phsubd ((__v2si)__X, (__v2si)__Y);
104 }
105
106 static __inline __m64 __attribute__((__always_inline__))
_mm_hsubs_pi16(__m64 __X,__m64 __Y)107 _mm_hsubs_pi16 (__m64 __X, __m64 __Y)
108 {
109 return (__m64) __builtin_ia32_phsubsw ((__v4hi)__X, (__v4hi)__Y);
110 }
111
112 static __inline __m128i __attribute__((__always_inline__))
_mm_maddubs_epi16(__m128i __X,__m128i __Y)113 _mm_maddubs_epi16 (__m128i __X, __m128i __Y)
114 {
115 return (__m128i) __builtin_ia32_pmaddubsw128 ((__v16qi)__X, (__v16qi)__Y);
116 }
117
118 static __inline __m64 __attribute__((__always_inline__))
_mm_maddubs_pi16(__m64 __X,__m64 __Y)119 _mm_maddubs_pi16 (__m64 __X, __m64 __Y)
120 {
121 return (__m64) __builtin_ia32_pmaddubsw ((__v8qi)__X, (__v8qi)__Y);
122 }
123
124 static __inline __m128i __attribute__((__always_inline__))
_mm_mulhrs_epi16(__m128i __X,__m128i __Y)125 _mm_mulhrs_epi16 (__m128i __X, __m128i __Y)
126 {
127 return (__m128i) __builtin_ia32_pmulhrsw128 ((__v8hi)__X, (__v8hi)__Y);
128 }
129
130 static __inline __m64 __attribute__((__always_inline__))
_mm_mulhrs_pi16(__m64 __X,__m64 __Y)131 _mm_mulhrs_pi16 (__m64 __X, __m64 __Y)
132 {
133 return (__m64) __builtin_ia32_pmulhrsw ((__v4hi)__X, (__v4hi)__Y);
134 }
135
136 static __inline __m128i __attribute__((__always_inline__))
_mm_shuffle_epi8(__m128i __X,__m128i __Y)137 _mm_shuffle_epi8 (__m128i __X, __m128i __Y)
138 {
139 return (__m128i) __builtin_ia32_pshufb128 ((__v16qi)__X, (__v16qi)__Y);
140 }
141
142 static __inline __m64 __attribute__((__always_inline__))
_mm_shuffle_pi8(__m64 __X,__m64 __Y)143 _mm_shuffle_pi8 (__m64 __X, __m64 __Y)
144 {
145 return (__m64) __builtin_ia32_pshufb ((__v8qi)__X, (__v8qi)__Y);
146 }
147
148 static __inline __m128i __attribute__((__always_inline__))
_mm_sign_epi8(__m128i __X,__m128i __Y)149 _mm_sign_epi8 (__m128i __X, __m128i __Y)
150 {
151 return (__m128i) __builtin_ia32_psignb128 ((__v16qi)__X, (__v16qi)__Y);
152 }
153
154 static __inline __m128i __attribute__((__always_inline__))
_mm_sign_epi16(__m128i __X,__m128i __Y)155 _mm_sign_epi16 (__m128i __X, __m128i __Y)
156 {
157 return (__m128i) __builtin_ia32_psignw128 ((__v8hi)__X, (__v8hi)__Y);
158 }
159
160 static __inline __m128i __attribute__((__always_inline__))
_mm_sign_epi32(__m128i __X,__m128i __Y)161 _mm_sign_epi32 (__m128i __X, __m128i __Y)
162 {
163 return (__m128i) __builtin_ia32_psignd128 ((__v4si)__X, (__v4si)__Y);
164 }
165
166 static __inline __m64 __attribute__((__always_inline__))
_mm_sign_pi8(__m64 __X,__m64 __Y)167 _mm_sign_pi8 (__m64 __X, __m64 __Y)
168 {
169 return (__m64) __builtin_ia32_psignb ((__v8qi)__X, (__v8qi)__Y);
170 }
171
172 static __inline __m64 __attribute__((__always_inline__))
_mm_sign_pi16(__m64 __X,__m64 __Y)173 _mm_sign_pi16 (__m64 __X, __m64 __Y)
174 {
175 return (__m64) __builtin_ia32_psignw ((__v4hi)__X, (__v4hi)__Y);
176 }
177
178 static __inline __m64 __attribute__((__always_inline__))
_mm_sign_pi32(__m64 __X,__m64 __Y)179 _mm_sign_pi32 (__m64 __X, __m64 __Y)
180 {
181 return (__m64) __builtin_ia32_psignd ((__v2si)__X, (__v2si)__Y);
182 }
183
184 #define _mm_alignr_epi8(__X, __Y, __N) \
185 ((__m128i)__builtin_ia32_palignr128 ((__v2di) __X, (__v2di) __Y, (__N) * 8))
186
187 #define _mm_alignr_pi8(__X, __Y, __N) \
188 ((__m64)__builtin_ia32_palignr ((long long) (__X), (long long) (__Y), (__N) * 8))
189
190 static __inline __m128i __attribute__((__always_inline__))
_mm_abs_epi8(__m128i __X)191 _mm_abs_epi8 (__m128i __X)
192 {
193 return (__m128i) __builtin_ia32_pabsb128 ((__v16qi)__X);
194 }
195
196 static __inline __m128i __attribute__((__always_inline__))
_mm_abs_epi16(__m128i __X)197 _mm_abs_epi16 (__m128i __X)
198 {
199 return (__m128i) __builtin_ia32_pabsw128 ((__v8hi)__X);
200 }
201
202 static __inline __m128i __attribute__((__always_inline__))
_mm_abs_epi32(__m128i __X)203 _mm_abs_epi32 (__m128i __X)
204 {
205 return (__m128i) __builtin_ia32_pabsd128 ((__v4si)__X);
206 }
207
208 static __inline __m64 __attribute__((__always_inline__))
_mm_abs_pi8(__m64 __X)209 _mm_abs_pi8 (__m64 __X)
210 {
211 return (__m64) __builtin_ia32_pabsb ((__v8qi)__X);
212 }
213
214 static __inline __m64 __attribute__((__always_inline__))
_mm_abs_pi16(__m64 __X)215 _mm_abs_pi16 (__m64 __X)
216 {
217 return (__m64) __builtin_ia32_pabsw ((__v4hi)__X);
218 }
219
220 static __inline __m64 __attribute__((__always_inline__))
_mm_abs_pi32(__m64 __X)221 _mm_abs_pi32 (__m64 __X)
222 {
223 return (__m64) __builtin_ia32_pabsd ((__v2si)__X);
224 }
225
226 #endif /* __SSSE3__ */
227
228 #endif /* _TMMINTRIN_H_INCLUDED */
229