Deleted Added
sdiff udiff text old ( 193576 ) new ( 198092 )
full compact
mmintrin.h (193576) mmintrin.h (198092)
1/*===---- mmintrin.h - MMX intrinsics --------------------------------------===
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to deal
5 * in the Software without restriction, including without limitation the rights
6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7 * copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19 * THE SOFTWARE.
20 *
21 *===-----------------------------------------------------------------------===
22 */
23
24#ifndef __MMINTRIN_H
25#define __MMINTRIN_H
26
27#ifndef __MMX__
28#error "MMX instruction set not enabled"
29#else
30
31typedef long long __m64 __attribute__((__vector_size__(8)));
32
33typedef int __v2si __attribute__((__vector_size__(8)));
34typedef short __v4hi __attribute__((__vector_size__(8)));
35typedef char __v8qi __attribute__((__vector_size__(8)));
36
37static inline void __attribute__((__always_inline__, __nodebug__))
38_mm_empty(void)
39{
40 __builtin_ia32_emms();
41}
42
43static inline __m64 __attribute__((__always_inline__, __nodebug__))
44_mm_cvtsi32_si64(int __i)
45{
46 return (__m64)(__v2si){__i, 0};
47}
48
49static inline int __attribute__((__always_inline__, __nodebug__))
50_mm_cvtsi64_si32(__m64 __m)
51{
52 __v2si __mmx_var2 = (__v2si)__m;
53 return __mmx_var2[0];
54}
55
56static inline __m64 __attribute__((__always_inline__, __nodebug__))
57_mm_cvtsi64_m64(long long __i)
58{
59 return (__m64)__i;
60}
61
62static inline long long __attribute__((__always_inline__, __nodebug__))
63_mm_cvtm64_si64(__m64 __m)
64{
65 return (long long)__m;
66}
67
68static inline __m64 __attribute__((__always_inline__, __nodebug__))
69_mm_packs_pi16(__m64 __m1, __m64 __m2)
70{
71 return (__m64)__builtin_ia32_packsswb((__v4hi)__m1, (__v4hi)__m2);
72}
73
74static inline __m64 __attribute__((__always_inline__, __nodebug__))
75_mm_packs_pi32(__m64 __m1, __m64 __m2)
76{
77 return (__m64)__builtin_ia32_packssdw((__v2si)__m1, (__v2si)__m2);
78}
79
80static inline __m64 __attribute__((__always_inline__, __nodebug__))
81_mm_packs_pu16(__m64 __m1, __m64 __m2)
82{
83 return (__m64)__builtin_ia32_packuswb((__v4hi)__m1, (__v4hi)__m2);
84}
85
86static inline __m64 __attribute__((__always_inline__, __nodebug__))
87_mm_unpackhi_pi8(__m64 __m1, __m64 __m2)
88{
89 return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 4, 8+4, 5,
90 8+5, 6, 8+6, 7, 8+7);
91}
92
93static inline __m64 __attribute__((__always_inline__, __nodebug__))
94_mm_unpackhi_pi16(__m64 __m1, __m64 __m2)
95{
96 return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 2, 4+2, 3,
97 4+3);
98}
99
100static inline __m64 __attribute__((__always_inline__, __nodebug__))
101_mm_unpackhi_pi32(__m64 __m1, __m64 __m2)
102{
103 return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 1, 2+1);
104}
105
106static inline __m64 __attribute__((__always_inline__, __nodebug__))
107_mm_unpacklo_pi8(__m64 __m1, __m64 __m2)
108{
109 return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 0, 8+0, 1,
110 8+1, 2, 8+2, 3, 8+3);
111}
112
113static inline __m64 __attribute__((__always_inline__, __nodebug__))
114_mm_unpacklo_pi16(__m64 __m1, __m64 __m2)
115{
116 return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 0, 4+0, 1,
117 4+1);
118}
119
120static inline __m64 __attribute__((__always_inline__, __nodebug__))
121_mm_unpacklo_pi32(__m64 __m1, __m64 __m2)
122{
123 return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 0, 2+0);
124}
125
126static inline __m64 __attribute__((__always_inline__, __nodebug__))
127_mm_add_pi8(__m64 __m1, __m64 __m2)
128{
129 return (__m64)((__v8qi)__m1 + (__v8qi)__m2);
130}
131
132static inline __m64 __attribute__((__always_inline__, __nodebug__))
133_mm_add_pi16(__m64 __m1, __m64 __m2)
134{
135 return (__m64)((__v4hi)__m1 + (__v4hi)__m2);
136}
137
138static inline __m64 __attribute__((__always_inline__, __nodebug__))
139_mm_add_pi32(__m64 __m1, __m64 __m2)
140{
141 return (__m64)((__v2si)__m1 + (__v2si)__m2);
142}
143
144static inline __m64 __attribute__((__always_inline__, __nodebug__))
145_mm_adds_pi8(__m64 __m1, __m64 __m2)
146{
147 return (__m64)__builtin_ia32_paddsb((__v8qi)__m1, (__v8qi)__m2);
148}
149
150static inline __m64 __attribute__((__always_inline__, __nodebug__))
151_mm_adds_pi16(__m64 __m1, __m64 __m2)
152{
153 return (__m64)__builtin_ia32_paddsw((__v4hi)__m1, (__v4hi)__m2);
154}
155
156static inline __m64 __attribute__((__always_inline__, __nodebug__))
157_mm_adds_pu8(__m64 __m1, __m64 __m2)
158{
159 return (__m64)__builtin_ia32_paddusb((__v8qi)__m1, (__v8qi)__m2);
160}
161
162static inline __m64 __attribute__((__always_inline__, __nodebug__))
163_mm_adds_pu16(__m64 __m1, __m64 __m2)
164{
165 return (__m64)__builtin_ia32_paddusw((__v4hi)__m1, (__v4hi)__m2);
166}
167
168static inline __m64 __attribute__((__always_inline__, __nodebug__))
169_mm_sub_pi8(__m64 __m1, __m64 __m2)
170{
171 return (__m64)((__v8qi)__m1 - (__v8qi)__m2);
172}
173
174static inline __m64 __attribute__((__always_inline__, __nodebug__))
175_mm_sub_pi16(__m64 __m1, __m64 __m2)
176{
177 return (__m64)((__v4hi)__m1 - (__v4hi)__m2);
178}
179
180static inline __m64 __attribute__((__always_inline__, __nodebug__))
181_mm_sub_pi32(__m64 __m1, __m64 __m2)
182{
183 return (__m64)((__v2si)__m1 - (__v2si)__m2);
184}
185
186static inline __m64 __attribute__((__always_inline__, __nodebug__))
187_mm_subs_pi8(__m64 __m1, __m64 __m2)
188{
189 return (__m64)__builtin_ia32_psubsb((__v8qi)__m1, (__v8qi)__m2);
190}
191
192static inline __m64 __attribute__((__always_inline__, __nodebug__))
193_mm_subs_pi16(__m64 __m1, __m64 __m2)
194{
195 return (__m64)__builtin_ia32_psubsw((__v4hi)__m1, (__v4hi)__m2);
196}
197
198static inline __m64 __attribute__((__always_inline__, __nodebug__))
199_mm_subs_pu8(__m64 __m1, __m64 __m2)
200{
201 return (__m64)__builtin_ia32_psubusb((__v8qi)__m1, (__v8qi)__m2);
202}
203
204static inline __m64 __attribute__((__always_inline__, __nodebug__))
205_mm_subs_pu16(__m64 __m1, __m64 __m2)
206{
207 return (__m64)__builtin_ia32_psubusw((__v4hi)__m1, (__v4hi)__m2);
208}
209
210static inline __m64 __attribute__((__always_inline__, __nodebug__))
211_mm_madd_pi16(__m64 __m1, __m64 __m2)
212{
213 return (__m64)__builtin_ia32_pmaddwd((__v4hi)__m1, (__v4hi)__m2);
214}
215
216static inline __m64 __attribute__((__always_inline__, __nodebug__))
217_mm_mulhi_pi16(__m64 __m1, __m64 __m2)
218{
219 return (__m64)__builtin_ia32_pmulhw((__v4hi)__m1, (__v4hi)__m2);
220}
221
222static inline __m64 __attribute__((__always_inline__, __nodebug__))
223_mm_mullo_pi16(__m64 __m1, __m64 __m2)
224{
225 return (__m64)((__v4hi)__m1 * (__v4hi)__m2);
226}
227
228static inline __m64 __attribute__((__always_inline__, __nodebug__))
229_mm_sll_pi16(__m64 __m, __m64 __count)
230{
231 return (__m64)__builtin_ia32_psllw((__v4hi)__m, __count);
232}
233
234static inline __m64 __attribute__((__always_inline__, __nodebug__))
235_mm_slli_pi16(__m64 __m, int __count)
236{
237 return (__m64)__builtin_ia32_psllwi((__v4hi)__m, __count);
238}
239
240static inline __m64 __attribute__((__always_inline__, __nodebug__))
241_mm_sll_pi32(__m64 __m, __m64 __count)
242{
243 return (__m64)__builtin_ia32_pslld((__v2si)__m, __count);
244}
245
246static inline __m64 __attribute__((__always_inline__, __nodebug__))
247_mm_slli_pi32(__m64 __m, int __count)
248{
249 return (__m64)__builtin_ia32_pslldi((__v2si)__m, __count);
250}
251
252static inline __m64 __attribute__((__always_inline__, __nodebug__))
253_mm_sll_si64(__m64 __m, __m64 __count)
254{
255 return __builtin_ia32_psllq(__m, __count);
256}
257
258static inline __m64 __attribute__((__always_inline__, __nodebug__))
259_mm_slli_si64(__m64 __m, int __count)
260{
261 return __builtin_ia32_psllqi(__m, __count);
262}
263
264static inline __m64 __attribute__((__always_inline__, __nodebug__))
265_mm_sra_pi16(__m64 __m, __m64 __count)
266{
267 return (__m64)__builtin_ia32_psraw((__v4hi)__m, __count);
268}
269
270static inline __m64 __attribute__((__always_inline__, __nodebug__))
271_mm_srai_pi16(__m64 __m, int __count)
272{
273 return (__m64)__builtin_ia32_psrawi((__v4hi)__m, __count);
274}
275
276static inline __m64 __attribute__((__always_inline__, __nodebug__))
277_mm_sra_pi32(__m64 __m, __m64 __count)
278{
279 return (__m64)__builtin_ia32_psrad((__v2si)__m, __count);
280}
281
282static inline __m64 __attribute__((__always_inline__, __nodebug__))
283_mm_srai_pi32(__m64 __m, int __count)
284{
285 return (__m64)__builtin_ia32_psradi((__v2si)__m, __count);
286}
287
288static inline __m64 __attribute__((__always_inline__, __nodebug__))
289_mm_srl_pi16(__m64 __m, __m64 __count)
290{
291 return (__m64)__builtin_ia32_psrlw((__v4hi)__m, __count);
292}
293
294static inline __m64 __attribute__((__always_inline__, __nodebug__))
295_mm_srli_pi16(__m64 __m, int __count)
296{
297 return (__m64)__builtin_ia32_psrlwi((__v4hi)__m, __count);
298}
299
300static inline __m64 __attribute__((__always_inline__, __nodebug__))
301_mm_srl_pi32(__m64 __m, __m64 __count)
302{
303 return (__m64)__builtin_ia32_psrld((__v2si)__m, __count);
304}
305
306static inline __m64 __attribute__((__always_inline__, __nodebug__))
307_mm_srli_pi32(__m64 __m, int __count)
308{
309 return (__m64)__builtin_ia32_psrldi((__v2si)__m, __count);
310}
311
312static inline __m64 __attribute__((__always_inline__, __nodebug__))
313_mm_srl_si64(__m64 __m, __m64 __count)
314{
315 return (__m64)__builtin_ia32_psrlq(__m, __count);
316}
317
318static inline __m64 __attribute__((__always_inline__, __nodebug__))
319_mm_srli_si64(__m64 __m, int __count)
320{
321 return __builtin_ia32_psrlqi(__m, __count);
322}
323
324static inline __m64 __attribute__((__always_inline__, __nodebug__))
325_mm_and_si64(__m64 __m1, __m64 __m2)
326{
327 return __m1 & __m2;
328}
329
330static inline __m64 __attribute__((__always_inline__, __nodebug__))
331_mm_andnot_si64(__m64 __m1, __m64 __m2)
332{
333 return ~__m1 & __m2;
334}
335
336static inline __m64 __attribute__((__always_inline__, __nodebug__))
337_mm_or_si64(__m64 __m1, __m64 __m2)
338{
339 return __m1 | __m2;
340}
341
342static inline __m64 __attribute__((__always_inline__, __nodebug__))
343_mm_xor_si64(__m64 __m1, __m64 __m2)
344{
345 return __m1 ^ __m2;
346}
347
348static inline __m64 __attribute__((__always_inline__, __nodebug__))
349_mm_cmpeq_pi8(__m64 __m1, __m64 __m2)
350{
1/*===---- mmintrin.h - MMX intrinsics --------------------------------------===
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to deal
5 * in the Software without restriction, including without limitation the rights
6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7 * copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19 * THE SOFTWARE.
20 *
21 *===-----------------------------------------------------------------------===
22 */
23
24#ifndef __MMINTRIN_H
25#define __MMINTRIN_H
26
27#ifndef __MMX__
28#error "MMX instruction set not enabled"
29#else
30
31typedef long long __m64 __attribute__((__vector_size__(8)));
32
33typedef int __v2si __attribute__((__vector_size__(8)));
34typedef short __v4hi __attribute__((__vector_size__(8)));
35typedef char __v8qi __attribute__((__vector_size__(8)));
36
37static inline void __attribute__((__always_inline__, __nodebug__))
38_mm_empty(void)
39{
40 __builtin_ia32_emms();
41}
42
43static inline __m64 __attribute__((__always_inline__, __nodebug__))
44_mm_cvtsi32_si64(int __i)
45{
46 return (__m64)(__v2si){__i, 0};
47}
48
49static inline int __attribute__((__always_inline__, __nodebug__))
50_mm_cvtsi64_si32(__m64 __m)
51{
52 __v2si __mmx_var2 = (__v2si)__m;
53 return __mmx_var2[0];
54}
55
56static inline __m64 __attribute__((__always_inline__, __nodebug__))
57_mm_cvtsi64_m64(long long __i)
58{
59 return (__m64)__i;
60}
61
62static inline long long __attribute__((__always_inline__, __nodebug__))
63_mm_cvtm64_si64(__m64 __m)
64{
65 return (long long)__m;
66}
67
68static inline __m64 __attribute__((__always_inline__, __nodebug__))
69_mm_packs_pi16(__m64 __m1, __m64 __m2)
70{
71 return (__m64)__builtin_ia32_packsswb((__v4hi)__m1, (__v4hi)__m2);
72}
73
74static inline __m64 __attribute__((__always_inline__, __nodebug__))
75_mm_packs_pi32(__m64 __m1, __m64 __m2)
76{
77 return (__m64)__builtin_ia32_packssdw((__v2si)__m1, (__v2si)__m2);
78}
79
80static inline __m64 __attribute__((__always_inline__, __nodebug__))
81_mm_packs_pu16(__m64 __m1, __m64 __m2)
82{
83 return (__m64)__builtin_ia32_packuswb((__v4hi)__m1, (__v4hi)__m2);
84}
85
86static inline __m64 __attribute__((__always_inline__, __nodebug__))
87_mm_unpackhi_pi8(__m64 __m1, __m64 __m2)
88{
89 return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 4, 8+4, 5,
90 8+5, 6, 8+6, 7, 8+7);
91}
92
93static inline __m64 __attribute__((__always_inline__, __nodebug__))
94_mm_unpackhi_pi16(__m64 __m1, __m64 __m2)
95{
96 return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 2, 4+2, 3,
97 4+3);
98}
99
100static inline __m64 __attribute__((__always_inline__, __nodebug__))
101_mm_unpackhi_pi32(__m64 __m1, __m64 __m2)
102{
103 return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 1, 2+1);
104}
105
106static inline __m64 __attribute__((__always_inline__, __nodebug__))
107_mm_unpacklo_pi8(__m64 __m1, __m64 __m2)
108{
109 return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 0, 8+0, 1,
110 8+1, 2, 8+2, 3, 8+3);
111}
112
113static inline __m64 __attribute__((__always_inline__, __nodebug__))
114_mm_unpacklo_pi16(__m64 __m1, __m64 __m2)
115{
116 return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 0, 4+0, 1,
117 4+1);
118}
119
120static inline __m64 __attribute__((__always_inline__, __nodebug__))
121_mm_unpacklo_pi32(__m64 __m1, __m64 __m2)
122{
123 return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 0, 2+0);
124}
125
126static inline __m64 __attribute__((__always_inline__, __nodebug__))
127_mm_add_pi8(__m64 __m1, __m64 __m2)
128{
129 return (__m64)((__v8qi)__m1 + (__v8qi)__m2);
130}
131
132static inline __m64 __attribute__((__always_inline__, __nodebug__))
133_mm_add_pi16(__m64 __m1, __m64 __m2)
134{
135 return (__m64)((__v4hi)__m1 + (__v4hi)__m2);
136}
137
138static inline __m64 __attribute__((__always_inline__, __nodebug__))
139_mm_add_pi32(__m64 __m1, __m64 __m2)
140{
141 return (__m64)((__v2si)__m1 + (__v2si)__m2);
142}
143
144static inline __m64 __attribute__((__always_inline__, __nodebug__))
145_mm_adds_pi8(__m64 __m1, __m64 __m2)
146{
147 return (__m64)__builtin_ia32_paddsb((__v8qi)__m1, (__v8qi)__m2);
148}
149
150static inline __m64 __attribute__((__always_inline__, __nodebug__))
151_mm_adds_pi16(__m64 __m1, __m64 __m2)
152{
153 return (__m64)__builtin_ia32_paddsw((__v4hi)__m1, (__v4hi)__m2);
154}
155
156static inline __m64 __attribute__((__always_inline__, __nodebug__))
157_mm_adds_pu8(__m64 __m1, __m64 __m2)
158{
159 return (__m64)__builtin_ia32_paddusb((__v8qi)__m1, (__v8qi)__m2);
160}
161
162static inline __m64 __attribute__((__always_inline__, __nodebug__))
163_mm_adds_pu16(__m64 __m1, __m64 __m2)
164{
165 return (__m64)__builtin_ia32_paddusw((__v4hi)__m1, (__v4hi)__m2);
166}
167
168static inline __m64 __attribute__((__always_inline__, __nodebug__))
169_mm_sub_pi8(__m64 __m1, __m64 __m2)
170{
171 return (__m64)((__v8qi)__m1 - (__v8qi)__m2);
172}
173
174static inline __m64 __attribute__((__always_inline__, __nodebug__))
175_mm_sub_pi16(__m64 __m1, __m64 __m2)
176{
177 return (__m64)((__v4hi)__m1 - (__v4hi)__m2);
178}
179
180static inline __m64 __attribute__((__always_inline__, __nodebug__))
181_mm_sub_pi32(__m64 __m1, __m64 __m2)
182{
183 return (__m64)((__v2si)__m1 - (__v2si)__m2);
184}
185
186static inline __m64 __attribute__((__always_inline__, __nodebug__))
187_mm_subs_pi8(__m64 __m1, __m64 __m2)
188{
189 return (__m64)__builtin_ia32_psubsb((__v8qi)__m1, (__v8qi)__m2);
190}
191
192static inline __m64 __attribute__((__always_inline__, __nodebug__))
193_mm_subs_pi16(__m64 __m1, __m64 __m2)
194{
195 return (__m64)__builtin_ia32_psubsw((__v4hi)__m1, (__v4hi)__m2);
196}
197
198static inline __m64 __attribute__((__always_inline__, __nodebug__))
199_mm_subs_pu8(__m64 __m1, __m64 __m2)
200{
201 return (__m64)__builtin_ia32_psubusb((__v8qi)__m1, (__v8qi)__m2);
202}
203
204static inline __m64 __attribute__((__always_inline__, __nodebug__))
205_mm_subs_pu16(__m64 __m1, __m64 __m2)
206{
207 return (__m64)__builtin_ia32_psubusw((__v4hi)__m1, (__v4hi)__m2);
208}
209
210static inline __m64 __attribute__((__always_inline__, __nodebug__))
211_mm_madd_pi16(__m64 __m1, __m64 __m2)
212{
213 return (__m64)__builtin_ia32_pmaddwd((__v4hi)__m1, (__v4hi)__m2);
214}
215
216static inline __m64 __attribute__((__always_inline__, __nodebug__))
217_mm_mulhi_pi16(__m64 __m1, __m64 __m2)
218{
219 return (__m64)__builtin_ia32_pmulhw((__v4hi)__m1, (__v4hi)__m2);
220}
221
222static inline __m64 __attribute__((__always_inline__, __nodebug__))
223_mm_mullo_pi16(__m64 __m1, __m64 __m2)
224{
225 return (__m64)((__v4hi)__m1 * (__v4hi)__m2);
226}
227
228static inline __m64 __attribute__((__always_inline__, __nodebug__))
229_mm_sll_pi16(__m64 __m, __m64 __count)
230{
231 return (__m64)__builtin_ia32_psllw((__v4hi)__m, __count);
232}
233
234static inline __m64 __attribute__((__always_inline__, __nodebug__))
235_mm_slli_pi16(__m64 __m, int __count)
236{
237 return (__m64)__builtin_ia32_psllwi((__v4hi)__m, __count);
238}
239
240static inline __m64 __attribute__((__always_inline__, __nodebug__))
241_mm_sll_pi32(__m64 __m, __m64 __count)
242{
243 return (__m64)__builtin_ia32_pslld((__v2si)__m, __count);
244}
245
246static inline __m64 __attribute__((__always_inline__, __nodebug__))
247_mm_slli_pi32(__m64 __m, int __count)
248{
249 return (__m64)__builtin_ia32_pslldi((__v2si)__m, __count);
250}
251
252static inline __m64 __attribute__((__always_inline__, __nodebug__))
253_mm_sll_si64(__m64 __m, __m64 __count)
254{
255 return __builtin_ia32_psllq(__m, __count);
256}
257
258static inline __m64 __attribute__((__always_inline__, __nodebug__))
259_mm_slli_si64(__m64 __m, int __count)
260{
261 return __builtin_ia32_psllqi(__m, __count);
262}
263
264static inline __m64 __attribute__((__always_inline__, __nodebug__))
265_mm_sra_pi16(__m64 __m, __m64 __count)
266{
267 return (__m64)__builtin_ia32_psraw((__v4hi)__m, __count);
268}
269
270static inline __m64 __attribute__((__always_inline__, __nodebug__))
271_mm_srai_pi16(__m64 __m, int __count)
272{
273 return (__m64)__builtin_ia32_psrawi((__v4hi)__m, __count);
274}
275
276static inline __m64 __attribute__((__always_inline__, __nodebug__))
277_mm_sra_pi32(__m64 __m, __m64 __count)
278{
279 return (__m64)__builtin_ia32_psrad((__v2si)__m, __count);
280}
281
282static inline __m64 __attribute__((__always_inline__, __nodebug__))
283_mm_srai_pi32(__m64 __m, int __count)
284{
285 return (__m64)__builtin_ia32_psradi((__v2si)__m, __count);
286}
287
288static inline __m64 __attribute__((__always_inline__, __nodebug__))
289_mm_srl_pi16(__m64 __m, __m64 __count)
290{
291 return (__m64)__builtin_ia32_psrlw((__v4hi)__m, __count);
292}
293
294static inline __m64 __attribute__((__always_inline__, __nodebug__))
295_mm_srli_pi16(__m64 __m, int __count)
296{
297 return (__m64)__builtin_ia32_psrlwi((__v4hi)__m, __count);
298}
299
300static inline __m64 __attribute__((__always_inline__, __nodebug__))
301_mm_srl_pi32(__m64 __m, __m64 __count)
302{
303 return (__m64)__builtin_ia32_psrld((__v2si)__m, __count);
304}
305
306static inline __m64 __attribute__((__always_inline__, __nodebug__))
307_mm_srli_pi32(__m64 __m, int __count)
308{
309 return (__m64)__builtin_ia32_psrldi((__v2si)__m, __count);
310}
311
312static inline __m64 __attribute__((__always_inline__, __nodebug__))
313_mm_srl_si64(__m64 __m, __m64 __count)
314{
315 return (__m64)__builtin_ia32_psrlq(__m, __count);
316}
317
318static inline __m64 __attribute__((__always_inline__, __nodebug__))
319_mm_srli_si64(__m64 __m, int __count)
320{
321 return __builtin_ia32_psrlqi(__m, __count);
322}
323
324static inline __m64 __attribute__((__always_inline__, __nodebug__))
325_mm_and_si64(__m64 __m1, __m64 __m2)
326{
327 return __m1 & __m2;
328}
329
330static inline __m64 __attribute__((__always_inline__, __nodebug__))
331_mm_andnot_si64(__m64 __m1, __m64 __m2)
332{
333 return ~__m1 & __m2;
334}
335
336static inline __m64 __attribute__((__always_inline__, __nodebug__))
337_mm_or_si64(__m64 __m1, __m64 __m2)
338{
339 return __m1 | __m2;
340}
341
342static inline __m64 __attribute__((__always_inline__, __nodebug__))
343_mm_xor_si64(__m64 __m1, __m64 __m2)
344{
345 return __m1 ^ __m2;
346}
347
348static inline __m64 __attribute__((__always_inline__, __nodebug__))
349_mm_cmpeq_pi8(__m64 __m1, __m64 __m2)
350{
351 return (__m64)__builtin_ia32_pcmpeqb((__v8qi)__m1, (__v8qi)__m2);
351 return (__m64)((__v8qi)__m1 == (__v8qi)__m2);
352}
353
354static inline __m64 __attribute__((__always_inline__, __nodebug__))
355_mm_cmpeq_pi16(__m64 __m1, __m64 __m2)
356{
352}
353
354static inline __m64 __attribute__((__always_inline__, __nodebug__))
355_mm_cmpeq_pi16(__m64 __m1, __m64 __m2)
356{
357 return (__m64)__builtin_ia32_pcmpeqw((__v4hi)__m1, (__v4hi)__m2);
357 return (__m64)((__v4hi)__m1 == (__v4hi)__m2);
358}
359
360static inline __m64 __attribute__((__always_inline__, __nodebug__))
361_mm_cmpeq_pi32(__m64 __m1, __m64 __m2)
362{
358}
359
360static inline __m64 __attribute__((__always_inline__, __nodebug__))
361_mm_cmpeq_pi32(__m64 __m1, __m64 __m2)
362{
363 return (__m64)__builtin_ia32_pcmpeqd((__v2si)__m1, (__v2si)__m2);
363 return (__m64)((__v2si)__m1 == (__v2si)__m2);
364}
365
366static inline __m64 __attribute__((__always_inline__, __nodebug__))
367_mm_cmpgt_pi8(__m64 __m1, __m64 __m2)
368{
364}
365
366static inline __m64 __attribute__((__always_inline__, __nodebug__))
367_mm_cmpgt_pi8(__m64 __m1, __m64 __m2)
368{
369 return (__m64)__builtin_ia32_pcmpgtb((__v8qi)__m1, (__v8qi)__m2);
369 return (__m64)((__v8qi)__m1 > (__v8qi)__m2);
370}
371
372static inline __m64 __attribute__((__always_inline__, __nodebug__))
373_mm_cmpgt_pi16(__m64 __m1, __m64 __m2)
374{
370}
371
372static inline __m64 __attribute__((__always_inline__, __nodebug__))
373_mm_cmpgt_pi16(__m64 __m1, __m64 __m2)
374{
375 return (__m64)__builtin_ia32_pcmpgtw((__v4hi)__m1, (__v4hi)__m2);
375 return (__m64)((__v4hi)__m1 > (__v4hi)__m2);
376}
377
378static inline __m64 __attribute__((__always_inline__, __nodebug__))
379_mm_cmpgt_pi32(__m64 __m1, __m64 __m2)
380{
376}
377
378static inline __m64 __attribute__((__always_inline__, __nodebug__))
379_mm_cmpgt_pi32(__m64 __m1, __m64 __m2)
380{
381 return (__m64)__builtin_ia32_pcmpgtd((__v2si)__m1, (__v2si)__m2);
381 return (__m64)((__v2si)__m1 > (__v2si)__m2);
382}
383
384static inline __m64 __attribute__((__always_inline__, __nodebug__))
385_mm_setzero_si64(void)
386{
387 return (__m64){ 0LL };
388}
389
390static inline __m64 __attribute__((__always_inline__, __nodebug__))
391_mm_set_pi32(int __i1, int __i0)
392{
393 return (__m64)(__v2si){ __i0, __i1 };
394}
395
396static inline __m64 __attribute__((__always_inline__, __nodebug__))
397_mm_set_pi16(short __s3, short __s2, short __s1, short __s0)
398{
399 return (__m64)(__v4hi){ __s0, __s1, __s2, __s3 };
400}
401
402static inline __m64 __attribute__((__always_inline__, __nodebug__))
403_mm_set_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
404 char __b1, char __b0)
405{
406 return (__m64)(__v8qi){ __b0, __b1, __b2, __b3, __b4, __b5, __b6, __b7 };
407}
408
409static inline __m64 __attribute__((__always_inline__, __nodebug__))
410_mm_set1_pi32(int __i)
411{
412 return (__m64)(__v2si){ __i, __i };
413}
414
415static inline __m64 __attribute__((__always_inline__, __nodebug__))
416_mm_set1_pi16(short __s)
417{
418 return (__m64)(__v4hi){ __s, __s, __s, __s };
419}
420
421static inline __m64 __attribute__((__always_inline__, __nodebug__))
422_mm_set1_pi8(char __b)
423{
424 return (__m64)(__v8qi){ __b, __b, __b, __b, __b, __b, __b, __b };
425}
426
427static inline __m64 __attribute__((__always_inline__, __nodebug__))
428_mm_setr_pi32(int __i1, int __i0)
429{
430 return (__m64)(__v2si){ __i1, __i0 };
431}
432
433static inline __m64 __attribute__((__always_inline__, __nodebug__))
434_mm_setr_pi16(short __s3, short __s2, short __s1, short __s0)
435{
436 return (__m64)(__v4hi){ __s3, __s2, __s1, __s0 };
437}
438
439static inline __m64 __attribute__((__always_inline__, __nodebug__))
440_mm_setr_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
441 char __b1, char __b0)
442{
443 return (__m64)(__v8qi){ __b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0 };
444}
445
446#endif /* __MMX__ */
447
448#endif /* __MMINTRIN_H */
449
382}
383
384static inline __m64 __attribute__((__always_inline__, __nodebug__))
385_mm_setzero_si64(void)
386{
387 return (__m64){ 0LL };
388}
389
390static inline __m64 __attribute__((__always_inline__, __nodebug__))
391_mm_set_pi32(int __i1, int __i0)
392{
393 return (__m64)(__v2si){ __i0, __i1 };
394}
395
396static inline __m64 __attribute__((__always_inline__, __nodebug__))
397_mm_set_pi16(short __s3, short __s2, short __s1, short __s0)
398{
399 return (__m64)(__v4hi){ __s0, __s1, __s2, __s3 };
400}
401
402static inline __m64 __attribute__((__always_inline__, __nodebug__))
403_mm_set_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
404 char __b1, char __b0)
405{
406 return (__m64)(__v8qi){ __b0, __b1, __b2, __b3, __b4, __b5, __b6, __b7 };
407}
408
409static inline __m64 __attribute__((__always_inline__, __nodebug__))
410_mm_set1_pi32(int __i)
411{
412 return (__m64)(__v2si){ __i, __i };
413}
414
415static inline __m64 __attribute__((__always_inline__, __nodebug__))
416_mm_set1_pi16(short __s)
417{
418 return (__m64)(__v4hi){ __s, __s, __s, __s };
419}
420
421static inline __m64 __attribute__((__always_inline__, __nodebug__))
422_mm_set1_pi8(char __b)
423{
424 return (__m64)(__v8qi){ __b, __b, __b, __b, __b, __b, __b, __b };
425}
426
427static inline __m64 __attribute__((__always_inline__, __nodebug__))
428_mm_setr_pi32(int __i1, int __i0)
429{
430 return (__m64)(__v2si){ __i1, __i0 };
431}
432
433static inline __m64 __attribute__((__always_inline__, __nodebug__))
434_mm_setr_pi16(short __s3, short __s2, short __s1, short __s0)
435{
436 return (__m64)(__v4hi){ __s3, __s2, __s1, __s0 };
437}
438
439static inline __m64 __attribute__((__always_inline__, __nodebug__))
440_mm_setr_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
441 char __b1, char __b0)
442{
443 return (__m64)(__v8qi){ __b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0 };
444}
445
446#endif /* __MMX__ */
447
448#endif /* __MMINTRIN_H */
449