mmintrin.h revision edbd9289b71cd512ae2b3a9b46cdf55b5602aaba 
1/*===---- mmintrin.h - MMX intrinsics --------------------------------------===
2 *
3 * Copyright (c) 2008 Anders Carlsson
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a copy
6 * of this software and associated documentation files (the "Software"), to deal
7 * in the Software without restriction, including without limitation the rights
8 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9 * copies of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
21 * THE SOFTWARE.
22 *
23 *===-----------------------------------------------------------------------===
24 */
25
26#ifndef __MMINTRIN_H
27#define __MMINTRIN_H
28
29#ifndef __MMX__
30#error "MMX instruction set not enabled"
31#else
32
33typedef long long __m64 __attribute__((__vector_size__(8)));
34
35typedef int __v2si __attribute__((__vector_size__(8)));
36typedef short __v4hi __attribute__((__vector_size__(8)));
37typedef char __v8qi __attribute__((__vector_size__(8)));
38
39inline void __attribute__((__always_inline__)) _mm_empty()
40{
41    __builtin_ia32_emms();
42}
43
44inline __m64 __attribute__((__always_inline__)) _mm_cvtsi32_si64(int __i)
45{
46    return (__m64)(__v2si){__i, 0};
47}
48
49inline int __attribute__((__always_inline__)) _mm_cvtsi64_si32(__m64 __m)
50{
51    __v2si __mmx_var2 = (__v2si)__m;
52    return __mmx_var2[0];
53}
54
55inline __m64 __attribute__((__always_inline__)) _mm_cvtsi64_m64(long long __i)
56{
57    return (__m64)__i;
58}
59
60inline long long __attribute__((__always_inline__)) _mm_cvtm64_si64(__m64 __m)
61{
62    return (long long)__m;
63}
64
65inline __m64 __attribute__((__always_inline__)) _mm_packs_pi16(__m64 __m1, __m64 __m2)
66{
67    return (__m64)__builtin_ia32_packsswb((__v4hi)__m1, (__v4hi)__m2);
68}
69
70inline __m64 __attribute__((__always_inline__)) _mm_packs_pi32(__m64 __m1, __m64 __m2)
71{
72    return (__m64)__builtin_ia32_packssdw((__v2si)__m1, (__v2si)__m2);
73}
74
75inline __m64 __attribute__((__always_inline__)) _mm_packs_pu16(__m64 __m1, __m64 __m2)
76{
77    return (__m64)__builtin_ia32_packuswb((__v4hi)__m1, (__v4hi)__m2);
78}
79
80inline __m64 __attribute__((__always_inline__)) _mm_unpackhi_pi8(__m64 __m1, __m64 __m2)
81{
82    return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 4, 8+4, 5, 8+5, 6, 8+6, 7, 8+7);
83}
84
85inline __m64 __attribute__((__always_inline__)) _mm_unpackhi_pi16(__m64 __m1, __m64 __m2)
86{
87    return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 2, 4+2, 3, 4+3);
88}
89
90inline __m64 __attribute__((__always_inline__)) _mm_unpackhi_pi32(__m64 __m1, __m64 __m2)
91{
92    return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 1, 2+1);
93}
94
95inline __m64 __attribute__((__always_inline__)) _mm_unpacklo_pi8(__m64 __m1, __m64 __m2)
96{
97    return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 0, 8+0, 1, 8+1, 2, 8+2, 3, 8+3);
98}
99
100inline __m64 __attribute__((__always_inline__)) _mm_unpacklo_pi16(__m64 __m1, __m64 __m2)
101{
102    return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 0, 4+0, 1, 4+1);
103}
104
105inline __m64 __attribute__((__always_inline__)) _mm_unpacklo_pi32(__m64 __m1, __m64 __m2)
106{
107    return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 0, 2+0);
108}
109
110inline __m64 __attribute__((__always_inline__)) _mm_add_pi8(__m64 __m1, __m64 __m2)
111{
112    return (__m64)((__v8qi)__m1 + (__v8qi)__m2);
113}
114
115inline __m64 __attribute__((__always_inline__)) _mm_add_pi16(__m64 __m1, __m64 __m2)
116{
117    return (__m64)((__v4hi)__m1 + (__v4hi)__m2);
118}
119
120inline __m64 __attribute__((__always_inline__)) _mm_add_pi32(__m64 __m1, __m64 __m2)
121{
122    return (__m64)((__v2si)__m1 + (__v2si)__m2);
123}
124
125inline __m64 __attribute__((__always_inline__)) _mm_adds_pi8(__m64 __m1, __m64 __m2)
126{
127    return (__m64)__builtin_ia32_paddsb((__v8qi)__m1, (__v8qi)__m2);
128}
129
130inline __m64 __attribute__((__always_inline__)) _mm_adds_pi16(__m64 __m1, __m64 __m2)
131{
132    return (__m64)__builtin_ia32_paddsw((__v4hi)__m1, (__v4hi)__m2);
133}
134
135inline __m64 __attribute__((__always_inline__)) _mm_adds_pu8(__m64 __m1, __m64 __m2)
136{
137    return (__m64)__builtin_ia32_paddusb((__v8qi)__m1, (__v8qi)__m2);
138}
139
140inline __m64 __attribute__((__always_inline__)) _mm_adds_pu16(__m64 __m1, __m64 __m2)
141{
142    return (__m64)__builtin_ia32_paddusw((__v4hi)__m1, (__v4hi)__m2);
143}
144
145inline __m64 __attribute__((__always_inline__)) _mm_sub_pi8(__m64 __m1, __m64 __m2)
146{
147    return (__m64)((__v8qi)__m1 - (__v8qi)__m2);
148}
149
150inline __m64 __attribute__((__always_inline__)) _mm_sub_pi16(__m64 __m1, __m64 __m2)
151{
152    return (__m64)((__v4hi)__m1 - (__v4hi)__m2);
153}
154
155inline __m64 __attribute__((__always_inline__)) _mm_sub_pi32(__m64 __m1, __m64 __m2)
156{
157    return (__m64)((__v2si)__m1 - (__v2si)__m2);
158}
159
160inline __m64 __attribute__((__always_inline__)) _mm_subs_pi8(__m64 __m1, __m64 __m2)
161{
162    return (__m64)__builtin_ia32_psubsb((__v8qi)__m1, (__v8qi)__m2);
163}
164
165inline __m64 __attribute__((__always_inline__)) _mm_subs_pi16(__m64 __m1, __m64 __m2)
166{
167    return (__m64)__builtin_ia32_psubsw((__v4hi)__m1, (__v4hi)__m2);
168}
169
170inline __m64 __attribute__((__always_inline__)) _mm_subs_pu8(__m64 __m1, __m64 __m2)
171{
172    return (__m64)__builtin_ia32_psubusb((__v8qi)__m1, (__v8qi)__m2);
173}
174
175inline __m64 __attribute__((__always_inline__)) _mm_subs_pu16(__m64 __m1, __m64 __m2)
176{
177    return (__m64)__builtin_ia32_psubusw((__v4hi)__m1, (__v4hi)__m2);
178}
179
180inline __m64 __attribute__((__always_inline__)) _mm_madd_pi16(__m64 __m1, __m64 __m2)
181{
182    return (__m64)__builtin_ia32_pmaddwd((__v4hi)__m1, (__v4hi)__m2);
183}
184
185inline __m64 __attribute__((__always_inline__)) _mm_mulhi_pi16(__m64 __m1, __m64 __m2)
186{
187    return (__m64)__builtin_ia32_pmulhw((__v4hi)__m1, (__v4hi)__m2);
188}
189
190inline __m64 __attribute__((__always_inline__)) _mm_mullo_pi16(__m64 __m1, __m64 __m2)
191{
192    return (__m64)((__v4hi)__m1 * (__v4hi)__m2);
193}
194
195inline __m64 __attribute__((__always_inline__)) _mm_sll_pi16(__m64 __m, __m64 __count)
196{
197    return (__m64)__builtin_ia32_psllw((__v4hi)__m, __count);
198}
199
200inline __m64 __attribute__((__always_inline__)) _mm_slli_pi16(__m64 __m, int __count)
201{
202    return (__m64)__builtin_ia32_psllwi((__v4hi)__m, __count);
203}
204
205inline __m64 __attribute__((__always_inline__)) _mm_sll_pi32(__m64 __m, __m64 __count)
206{
207    return (__m64)__builtin_ia32_pslld((__v2si)__m, __count);
208}
209
210inline __m64 __attribute__((__always_inline__)) _mm_slli_pi32(__m64 __m, int __count)
211{
212    return (__m64)__builtin_ia32_pslldi((__v2si)__m, __count);
213}
214
215inline __m64 __attribute__((__always_inline__)) _mm_sll_pi64(__m64 __m, __m64 __count)
216{
217    return __builtin_ia32_psllq(__m, __count);
218}
219
220inline __m64 __attribute__((__always_inline__)) _mm_slli_pi64(__m64 __m, int __count)
221{
222    return __builtin_ia32_psllqi(__m, __count);
223}
224
225inline __m64 __attribute__((__always_inline__)) _mm_sra_pi16(__m64 __m, __m64 __count)
226{
227    return (__m64)__builtin_ia32_psraw((__v4hi)__m, __count);
228}
229
230inline __m64 __attribute__((__always_inline__)) _mm_srai_pi16(__m64 __m, int __count)
231{
232    return (__m64)__builtin_ia32_psrawi((__v4hi)__m, __count);
233}
234
235inline __m64 __attribute__((__always_inline__)) _mm_sra_pi32(__m64 __m, __m64 __count)
236{
237    return (__m64)__builtin_ia32_psrad((__v2si)__m, __count);
238}
239
240inline __m64 __attribute__((__always_inline__)) _mm_srai_pi32(__m64 __m, int __count)
241{
242    return (__m64)__builtin_ia32_psradi((__v2si)__m, __count);
243}
244
245inline __m64 __attribute__((__always_inline__)) _mm_srl_pi16(__m64 __m, __m64 __count)
246{
247    return (__m64)__builtin_ia32_psrlw((__v4hi)__m, __count);
248}
249
250inline __m64 __attribute__((__always_inline__)) _mm_srli_pi16(__m64 __m, int __count)
251{
252    return (__m64)__builtin_ia32_psrlwi((__v4hi)__m, __count);
253}
254
255inline __m64 __attribute__((__always_inline__)) _mm_srl_pi32(__m64 __m, __m64 __count)
256{
257    return (__m64)__builtin_ia32_psrld((__v2si)__m, __count);
258}
259
260inline __m64 __attribute__((__always_inline__)) _mm_srli_pi32(__m64 __m, int __count)
261{
262    return (__m64)__builtin_ia32_psrldi((__v2si)__m, __count);
263}
264
265inline __m64 __attribute__((__always_inline__)) _mm_srl_pi64(__m64 __m, __m64 __count)
266{
267    return (__m64)__builtin_ia32_psrlq(__m, __count);
268}
269
270inline __m64 __attribute__((__always_inline__)) _mm_srli_pi64(__m64 __m, int __count)
271{
272    return __builtin_ia32_psrlqi(__m, __count);
273}
274
275inline __m64 __attribute__((__always_inline__)) _mm_and_si64(__m64 __m1, __m64 __m2)
276{
277    return __m1 & __m2;
278}
279
280inline __m64 __attribute__((__always_inline__)) _mm_andnot_si64(__m64 __m1, __m64 __m2)
281{
282    return ~__m1 & __m2;
283}
284
285inline __m64 __attribute__((__always_inline__)) _mm_or_si64(__m64 __m1, __m64 __m2)
286{
287    return __m1 | __m2;
288}
289
290inline __m64 __attribute__((__always_inline__)) _mm_xor_si64(__m64 __m1, __m64 __m2)
291{
292    return __m1 ^ __m2;
293}
294
295inline __m64 __attribute__((__always_inline__)) _mm_cmpeq_pi8(__m64 __m1, __m64 __m2)
296{
297    return (__m64)__builtin_ia32_pcmpeqb((__v8qi)__m1, (__v8qi)__m2);
298}
299
300inline __m64 __attribute__((__always_inline__)) _mm_cmpeq_pi16(__m64 __m1, __m64 __m2)
301{
302    return (__m64)__builtin_ia32_pcmpeqw((__v4hi)__m1, (__v4hi)__m2);
303}
304
305inline __m64 __attribute__((__always_inline__)) _mm_cmpeq_pi32(__m64 __m1, __m64 __m2)
306{
307    return (__m64)__builtin_ia32_pcmpeqd((__v2si)__m1, (__v2si)__m2);
308}
309
310inline __m64 __attribute__((__always_inline__)) _mm_cmpgt_pi8(__m64 __m1, __m64 __m2)
311{
312    return (__m64)__builtin_ia32_pcmpgtb((__v8qi)__m1, (__v8qi)__m2);
313}
314
315inline __m64 __attribute__((__always_inline__)) _mm_cmpgt_pi16(__m64 __m1, __m64 __m2)
316{
317    return (__m64)__builtin_ia32_pcmpgtw((__v4hi)__m1, (__v4hi)__m2);
318}
319
320inline __m64 __attribute__((__always_inline__)) _mm_cmpgt_pi32(__m64 __m1, __m64 __m2)
321{
322    return (__m64)__builtin_ia32_pcmpgtd((__v2si)__m1, (__v2si)__m2);
323}
324
325inline __m64 __attribute__((__always_inline__)) _mm_setzero_si64()
326{
327    return (__m64){ 0LL };
328}
329
330inline __m64 __attribute__((__always_inline__)) _mm_set_pi32(int __i1, int __i0)
331{
332    return (__m64)(__v2si){ __i0, __i1 };
333}
334
335inline __m64 __attribute__((__always_inline__)) _mm_set_pi16(short __s3, short __s2, short __s1, short __s0)
336{
337    return (__m64)(__v4hi){ __s0, __s1, __s2, __s3 };
338}
339
340inline __m64 __attribute__((__always_inline__)) _mm_set_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2, char __b1, char __b0)
341{
342    return (__m64)(__v8qi){ __b0, __b1, __b2, __b3, __b4, __b5, __b6, __b7 };
343}
344
345inline __m64 __attribute__((__always_inline__)) _mm_set1_pi32(int __i)
346{
347    return (__m64)(__v2si){ __i, __i };
348}
349
350inline __m64 __attribute__((__always_inline__)) _mm_set1_pi16(short __s)
351{
352    return (__m64)(__v4hi){ __s };
353}
354
355inline __m64 __attribute__((__always_inline__)) _mm_set1_pi8(char __b)
356{
357    return (__m64)(__v8qi){ __b };
358}
359
360inline __m64 __attribute__((__always_inline__)) _mm_setr_pi32(int __i1, int __i0)
361{
362    return (__m64)(__v2si){ __i1, __i0 };
363}
364
365inline __m64 __attribute__((__always_inline__)) _mm_setr_pi16(short __s3, short __s2, short __s1, short __s0)
366{
367    return (__m64)(__v4hi){ __s3, __s2, __s1, __s0 };
368}
369
370inline __m64 __attribute__((__always_inline__)) _mm_setr_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2, char __b1, char __b0)
371{
372    return (__m64)(__v8qi){ __b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0 };
373}
374
375#endif /* __MMX__ */
376
377#endif /* __MMINTRIN_H */
378
379