Mercurial > libavcodec.hg

comparison x86/h264_qpel_mmx.c @ 12450:3941687b4fa9 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Split h264dsp_mmx.c (which was #included in dsputil_mmx.c) in h264_qpel_mmx.c, still #included in dsputil_mmx.c and is part of DSPContext, and h264dsp_mmx.c, which represents H264DSPContext and is now compiled on its own.
author rbultje
date Wed, 01 Sep 2010 20:48:59 +0000
parents
children
comparison
equal deleted inserted replaced
12449:3bca212d6f51 12450:3941687b4fa9
1 /*
2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 #include "dsputil_mmx.h"
22
23 /***********************************/
24 /* motion compensation */
25
26 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
27 "mov"#q" "#C", "#T" \n\t"\
28 "mov"#d" (%0), "#F" \n\t"\
29 "paddw "#D", "#T" \n\t"\
30 "psllw $2, "#T" \n\t"\
31 "psubw "#B", "#T" \n\t"\
32 "psubw "#E", "#T" \n\t"\
33 "punpcklbw "#Z", "#F" \n\t"\
34 "pmullw %4, "#T" \n\t"\
35 "paddw %5, "#A" \n\t"\
36 "add %2, %0 \n\t"\
37 "paddw "#F", "#A" \n\t"\
38 "paddw "#A", "#T" \n\t"\
39 "psraw $5, "#T" \n\t"\
40 "packuswb "#T", "#T" \n\t"\
41 OP(T, (%1), A, d)\
42 "add %3, %1 \n\t"
43
44 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
45 "mov"#q" "#C", "#T" \n\t"\
46 "mov"#d" (%0), "#F" \n\t"\
47 "paddw "#D", "#T" \n\t"\
48 "psllw $2, "#T" \n\t"\
49 "paddw %4, "#A" \n\t"\
50 "psubw "#B", "#T" \n\t"\
51 "psubw "#E", "#T" \n\t"\
52 "punpcklbw "#Z", "#F" \n\t"\
53 "pmullw %3, "#T" \n\t"\
54 "paddw "#F", "#A" \n\t"\
55 "add %2, %0 \n\t"\
56 "paddw "#A", "#T" \n\t"\
57 "mov"#q" "#T", "#OF"(%1) \n\t"
58
59 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
60 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
61 #define QPEL_H264V_XMM(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%xmm6,%%xmm7,q,dqa)
62 #define QPEL_H264HV_XMM(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%xmm6,%%xmm7,q,dqa)
63
64
65 #define QPEL_H264(OPNAME, OP, MMX)\
66 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
67 int h=4;\
68 \
69 __asm__ volatile(\
70 "pxor %%mm7, %%mm7 \n\t"\
71 "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
72 "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
73 "1: \n\t"\
74 "movd -1(%0), %%mm1 \n\t"\
75 "movd (%0), %%mm2 \n\t"\
76 "movd 1(%0), %%mm3 \n\t"\
77 "movd 2(%0), %%mm0 \n\t"\
78 "punpcklbw %%mm7, %%mm1 \n\t"\
79 "punpcklbw %%mm7, %%mm2 \n\t"\
80 "punpcklbw %%mm7, %%mm3 \n\t"\
81 "punpcklbw %%mm7, %%mm0 \n\t"\
82 "paddw %%mm0, %%mm1 \n\t"\
83 "paddw %%mm3, %%mm2 \n\t"\
84 "movd -2(%0), %%mm0 \n\t"\
85 "movd 3(%0), %%mm3 \n\t"\
86 "punpcklbw %%mm7, %%mm0 \n\t"\
87 "punpcklbw %%mm7, %%mm3 \n\t"\
88 "paddw %%mm3, %%mm0 \n\t"\
89 "psllw $2, %%mm2 \n\t"\
90 "psubw %%mm1, %%mm2 \n\t"\
91 "pmullw %%mm4, %%mm2 \n\t"\
92 "paddw %%mm5, %%mm0 \n\t"\
93 "paddw %%mm2, %%mm0 \n\t"\
94 "psraw $5, %%mm0 \n\t"\
95 "packuswb %%mm0, %%mm0 \n\t"\
96 OP(%%mm0, (%1),%%mm6, d)\
97 "add %3, %0 \n\t"\
98 "add %4, %1 \n\t"\
99 "decl %2 \n\t"\
100 " jnz 1b \n\t"\
101 : "+a"(src), "+c"(dst), "+g"(h)\
102 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
103 : "memory"\
104 );\
105 }\
106 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
107 int h=4;\
108 __asm__ volatile(\
109 "pxor %%mm7, %%mm7 \n\t"\
110 "movq %0, %%mm4 \n\t"\
111 "movq %1, %%mm5 \n\t"\
112 :: "m"(ff_pw_5), "m"(ff_pw_16)\
113 );\
114 do{\
115 __asm__ volatile(\
116 "movd -1(%0), %%mm1 \n\t"\
117 "movd (%0), %%mm2 \n\t"\
118 "movd 1(%0), %%mm3 \n\t"\
119 "movd 2(%0), %%mm0 \n\t"\
120 "punpcklbw %%mm7, %%mm1 \n\t"\
121 "punpcklbw %%mm7, %%mm2 \n\t"\
122 "punpcklbw %%mm7, %%mm3 \n\t"\
123 "punpcklbw %%mm7, %%mm0 \n\t"\
124 "paddw %%mm0, %%mm1 \n\t"\
125 "paddw %%mm3, %%mm2 \n\t"\
126 "movd -2(%0), %%mm0 \n\t"\
127 "movd 3(%0), %%mm3 \n\t"\
128 "punpcklbw %%mm7, %%mm0 \n\t"\
129 "punpcklbw %%mm7, %%mm3 \n\t"\
130 "paddw %%mm3, %%mm0 \n\t"\
131 "psllw $2, %%mm2 \n\t"\
132 "psubw %%mm1, %%mm2 \n\t"\
133 "pmullw %%mm4, %%mm2 \n\t"\
134 "paddw %%mm5, %%mm0 \n\t"\
135 "paddw %%mm2, %%mm0 \n\t"\
136 "movd (%2), %%mm3 \n\t"\
137 "psraw $5, %%mm0 \n\t"\
138 "packuswb %%mm0, %%mm0 \n\t"\
139 PAVGB" %%mm3, %%mm0 \n\t"\
140 OP(%%mm0, (%1),%%mm6, d)\
141 "add %4, %0 \n\t"\
142 "add %4, %1 \n\t"\
143 "add %3, %2 \n\t"\
144 : "+a"(src), "+c"(dst), "+d"(src2)\
145 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
146 : "memory"\
147 );\
148 }while(--h);\
149 }\
150 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
151 src -= 2*srcStride;\
152 __asm__ volatile(\
153 "pxor %%mm7, %%mm7 \n\t"\
154 "movd (%0), %%mm0 \n\t"\
155 "add %2, %0 \n\t"\
156 "movd (%0), %%mm1 \n\t"\
157 "add %2, %0 \n\t"\
158 "movd (%0), %%mm2 \n\t"\
159 "add %2, %0 \n\t"\
160 "movd (%0), %%mm3 \n\t"\
161 "add %2, %0 \n\t"\
162 "movd (%0), %%mm4 \n\t"\
163 "add %2, %0 \n\t"\
164 "punpcklbw %%mm7, %%mm0 \n\t"\
165 "punpcklbw %%mm7, %%mm1 \n\t"\
166 "punpcklbw %%mm7, %%mm2 \n\t"\
167 "punpcklbw %%mm7, %%mm3 \n\t"\
168 "punpcklbw %%mm7, %%mm4 \n\t"\
169 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
170 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
171 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
172 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
173 \
174 : "+a"(src), "+c"(dst)\
175 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
176 : "memory"\
177 );\
178 }\
179 static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
180 int h=4;\
181 int w=3;\
182 src -= 2*srcStride+2;\
183 while(w--){\
184 __asm__ volatile(\
185 "pxor %%mm7, %%mm7 \n\t"\
186 "movd (%0), %%mm0 \n\t"\
187 "add %2, %0 \n\t"\
188 "movd (%0), %%mm1 \n\t"\
189 "add %2, %0 \n\t"\
190 "movd (%0), %%mm2 \n\t"\
191 "add %2, %0 \n\t"\
192 "movd (%0), %%mm3 \n\t"\
193 "add %2, %0 \n\t"\
194 "movd (%0), %%mm4 \n\t"\
195 "add %2, %0 \n\t"\
196 "punpcklbw %%mm7, %%mm0 \n\t"\
197 "punpcklbw %%mm7, %%mm1 \n\t"\
198 "punpcklbw %%mm7, %%mm2 \n\t"\
199 "punpcklbw %%mm7, %%mm3 \n\t"\
200 "punpcklbw %%mm7, %%mm4 \n\t"\
201 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
202 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
203 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
204 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
205 \
206 : "+a"(src)\
207 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
208 : "memory"\
209 );\
210 tmp += 4;\
211 src += 4 - 9*srcStride;\
212 }\
213 tmp -= 3*4;\
214 __asm__ volatile(\
215 "1: \n\t"\
216 "movq (%0), %%mm0 \n\t"\
217 "paddw 10(%0), %%mm0 \n\t"\
218 "movq 2(%0), %%mm1 \n\t"\
219 "paddw 8(%0), %%mm1 \n\t"\
220 "movq 4(%0), %%mm2 \n\t"\
221 "paddw 6(%0), %%mm2 \n\t"\
222 "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
223 "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
224 "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
225 "paddsw %%mm2, %%mm0 \n\t"\
226 "psraw $2, %%mm0 \n\t"/*((a-b)/4-b+c)/4 */\
227 "paddw %%mm2, %%mm0 \n\t"/*(a-5*b+20*c)/16 */\
228 "psraw $6, %%mm0 \n\t"\
229 "packuswb %%mm0, %%mm0 \n\t"\
230 OP(%%mm0, (%1),%%mm7, d)\
231 "add $24, %0 \n\t"\
232 "add %3, %1 \n\t"\
233 "decl %2 \n\t"\
234 " jnz 1b \n\t"\
235 : "+a"(tmp), "+c"(dst), "+g"(h)\
236 : "S"((x86_reg)dstStride)\
237 : "memory"\
238 );\
239 }\
240 \
241 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
242 int h=8;\
243 __asm__ volatile(\
244 "pxor %%mm7, %%mm7 \n\t"\
245 "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
246 "1: \n\t"\
247 "movq (%0), %%mm0 \n\t"\
248 "movq 1(%0), %%mm2 \n\t"\
249 "movq %%mm0, %%mm1 \n\t"\
250 "movq %%mm2, %%mm3 \n\t"\
251 "punpcklbw %%mm7, %%mm0 \n\t"\
252 "punpckhbw %%mm7, %%mm1 \n\t"\
253 "punpcklbw %%mm7, %%mm2 \n\t"\
254 "punpckhbw %%mm7, %%mm3 \n\t"\
255 "paddw %%mm2, %%mm0 \n\t"\
256 "paddw %%mm3, %%mm1 \n\t"\
257 "psllw $2, %%mm0 \n\t"\
258 "psllw $2, %%mm1 \n\t"\
259 "movq -1(%0), %%mm2 \n\t"\
260 "movq 2(%0), %%mm4 \n\t"\
261 "movq %%mm2, %%mm3 \n\t"\
262 "movq %%mm4, %%mm5 \n\t"\
263 "punpcklbw %%mm7, %%mm2 \n\t"\
264 "punpckhbw %%mm7, %%mm3 \n\t"\
265 "punpcklbw %%mm7, %%mm4 \n\t"\
266 "punpckhbw %%mm7, %%mm5 \n\t"\
267 "paddw %%mm4, %%mm2 \n\t"\
268 "paddw %%mm3, %%mm5 \n\t"\
269 "psubw %%mm2, %%mm0 \n\t"\
270 "psubw %%mm5, %%mm1 \n\t"\
271 "pmullw %%mm6, %%mm0 \n\t"\
272 "pmullw %%mm6, %%mm1 \n\t"\
273 "movd -2(%0), %%mm2 \n\t"\
274 "movd 7(%0), %%mm5 \n\t"\
275 "punpcklbw %%mm7, %%mm2 \n\t"\
276 "punpcklbw %%mm7, %%mm5 \n\t"\
277 "paddw %%mm3, %%mm2 \n\t"\
278 "paddw %%mm5, %%mm4 \n\t"\
279 "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
280 "paddw %%mm5, %%mm2 \n\t"\
281 "paddw %%mm5, %%mm4 \n\t"\
282 "paddw %%mm2, %%mm0 \n\t"\
283 "paddw %%mm4, %%mm1 \n\t"\
284 "psraw $5, %%mm0 \n\t"\
285 "psraw $5, %%mm1 \n\t"\
286 "packuswb %%mm1, %%mm0 \n\t"\
287 OP(%%mm0, (%1),%%mm5, q)\
288 "add %3, %0 \n\t"\
289 "add %4, %1 \n\t"\
290 "decl %2 \n\t"\
291 " jnz 1b \n\t"\
292 : "+a"(src), "+c"(dst), "+g"(h)\
293 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
294 : "memory"\
295 );\
296 }\
297 \
298 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
299 int h=8;\
300 __asm__ volatile(\
301 "pxor %%mm7, %%mm7 \n\t"\
302 "movq %0, %%mm6 \n\t"\
303 :: "m"(ff_pw_5)\
304 );\
305 do{\
306 __asm__ volatile(\
307 "movq (%0), %%mm0 \n\t"\
308 "movq 1(%0), %%mm2 \n\t"\
309 "movq %%mm0, %%mm1 \n\t"\
310 "movq %%mm2, %%mm3 \n\t"\
311 "punpcklbw %%mm7, %%mm0 \n\t"\
312 "punpckhbw %%mm7, %%mm1 \n\t"\
313 "punpcklbw %%mm7, %%mm2 \n\t"\
314 "punpckhbw %%mm7, %%mm3 \n\t"\
315 "paddw %%mm2, %%mm0 \n\t"\
316 "paddw %%mm3, %%mm1 \n\t"\
317 "psllw $2, %%mm0 \n\t"\
318 "psllw $2, %%mm1 \n\t"\
319 "movq -1(%0), %%mm2 \n\t"\
320 "movq 2(%0), %%mm4 \n\t"\
321 "movq %%mm2, %%mm3 \n\t"\
322 "movq %%mm4, %%mm5 \n\t"\
323 "punpcklbw %%mm7, %%mm2 \n\t"\
324 "punpckhbw %%mm7, %%mm3 \n\t"\
325 "punpcklbw %%mm7, %%mm4 \n\t"\
326 "punpckhbw %%mm7, %%mm5 \n\t"\
327 "paddw %%mm4, %%mm2 \n\t"\
328 "paddw %%mm3, %%mm5 \n\t"\
329 "psubw %%mm2, %%mm0 \n\t"\
330 "psubw %%mm5, %%mm1 \n\t"\
331 "pmullw %%mm6, %%mm0 \n\t"\
332 "pmullw %%mm6, %%mm1 \n\t"\
333 "movd -2(%0), %%mm2 \n\t"\
334 "movd 7(%0), %%mm5 \n\t"\
335 "punpcklbw %%mm7, %%mm2 \n\t"\
336 "punpcklbw %%mm7, %%mm5 \n\t"\
337 "paddw %%mm3, %%mm2 \n\t"\
338 "paddw %%mm5, %%mm4 \n\t"\
339 "movq %5, %%mm5 \n\t"\
340 "paddw %%mm5, %%mm2 \n\t"\
341 "paddw %%mm5, %%mm4 \n\t"\
342 "paddw %%mm2, %%mm0 \n\t"\
343 "paddw %%mm4, %%mm1 \n\t"\
344 "psraw $5, %%mm0 \n\t"\
345 "psraw $5, %%mm1 \n\t"\
346 "movq (%2), %%mm4 \n\t"\
347 "packuswb %%mm1, %%mm0 \n\t"\
348 PAVGB" %%mm4, %%mm0 \n\t"\
349 OP(%%mm0, (%1),%%mm5, q)\
350 "add %4, %0 \n\t"\
351 "add %4, %1 \n\t"\
352 "add %3, %2 \n\t"\
353 : "+a"(src), "+c"(dst), "+d"(src2)\
354 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
355 "m"(ff_pw_16)\
356 : "memory"\
357 );\
358 }while(--h);\
359 }\
360 \
361 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
362 int w= 2;\
363 src -= 2*srcStride;\
364 \
365 while(w--){\
366 __asm__ volatile(\
367 "pxor %%mm7, %%mm7 \n\t"\
368 "movd (%0), %%mm0 \n\t"\
369 "add %2, %0 \n\t"\
370 "movd (%0), %%mm1 \n\t"\
371 "add %2, %0 \n\t"\
372 "movd (%0), %%mm2 \n\t"\
373 "add %2, %0 \n\t"\
374 "movd (%0), %%mm3 \n\t"\
375 "add %2, %0 \n\t"\
376 "movd (%0), %%mm4 \n\t"\
377 "add %2, %0 \n\t"\
378 "punpcklbw %%mm7, %%mm0 \n\t"\
379 "punpcklbw %%mm7, %%mm1 \n\t"\
380 "punpcklbw %%mm7, %%mm2 \n\t"\
381 "punpcklbw %%mm7, %%mm3 \n\t"\
382 "punpcklbw %%mm7, %%mm4 \n\t"\
383 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
384 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
385 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
386 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
387 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
388 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
389 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
390 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
391 \
392 : "+a"(src), "+c"(dst)\
393 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
394 : "memory"\
395 );\
396 if(h==16){\
397 __asm__ volatile(\
398 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
399 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
400 QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
401 QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
402 QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
403 QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
404 QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
405 QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
406 \
407 : "+a"(src), "+c"(dst)\
408 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
409 : "memory"\
410 );\
411 }\
412 src += 4-(h+5)*srcStride;\
413 dst += 4-h*dstStride;\
414 }\
415 }\
416 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
417 int w = (size+8)>>2;\
418 src -= 2*srcStride+2;\
419 while(w--){\
420 __asm__ volatile(\
421 "pxor %%mm7, %%mm7 \n\t"\
422 "movd (%0), %%mm0 \n\t"\
423 "add %2, %0 \n\t"\
424 "movd (%0), %%mm1 \n\t"\
425 "add %2, %0 \n\t"\
426 "movd (%0), %%mm2 \n\t"\
427 "add %2, %0 \n\t"\
428 "movd (%0), %%mm3 \n\t"\
429 "add %2, %0 \n\t"\
430 "movd (%0), %%mm4 \n\t"\
431 "add %2, %0 \n\t"\
432 "punpcklbw %%mm7, %%mm0 \n\t"\
433 "punpcklbw %%mm7, %%mm1 \n\t"\
434 "punpcklbw %%mm7, %%mm2 \n\t"\
435 "punpcklbw %%mm7, %%mm3 \n\t"\
436 "punpcklbw %%mm7, %%mm4 \n\t"\
437 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
438 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
439 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
440 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
441 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
442 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
443 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
444 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
445 : "+a"(src)\
446 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
447 : "memory"\
448 );\
449 if(size==16){\
450 __asm__ volatile(\
451 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 8*48)\
452 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 9*48)\
453 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
454 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
455 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
456 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
457 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
458 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
459 : "+a"(src)\
460 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
461 : "memory"\
462 );\
463 }\
464 tmp += 4;\
465 src += 4 - (size+5)*srcStride;\
466 }\
467 }\
468 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
469 int w = size>>4;\
470 do{\
471 int h = size;\
472 __asm__ volatile(\
473 "1: \n\t"\
474 "movq (%0), %%mm0 \n\t"\
475 "movq 8(%0), %%mm3 \n\t"\
476 "movq 2(%0), %%mm1 \n\t"\
477 "movq 10(%0), %%mm4 \n\t"\
478 "paddw %%mm4, %%mm0 \n\t"\
479 "paddw %%mm3, %%mm1 \n\t"\
480 "paddw 18(%0), %%mm3 \n\t"\
481 "paddw 16(%0), %%mm4 \n\t"\
482 "movq 4(%0), %%mm2 \n\t"\
483 "movq 12(%0), %%mm5 \n\t"\
484 "paddw 6(%0), %%mm2 \n\t"\
485 "paddw 14(%0), %%mm5 \n\t"\
486 "psubw %%mm1, %%mm0 \n\t"\
487 "psubw %%mm4, %%mm3 \n\t"\
488 "psraw $2, %%mm0 \n\t"\
489 "psraw $2, %%mm3 \n\t"\
490 "psubw %%mm1, %%mm0 \n\t"\
491 "psubw %%mm4, %%mm3 \n\t"\
492 "paddsw %%mm2, %%mm0 \n\t"\
493 "paddsw %%mm5, %%mm3 \n\t"\
494 "psraw $2, %%mm0 \n\t"\
495 "psraw $2, %%mm3 \n\t"\
496 "paddw %%mm2, %%mm0 \n\t"\
497 "paddw %%mm5, %%mm3 \n\t"\
498 "psraw $6, %%mm0 \n\t"\
499 "psraw $6, %%mm3 \n\t"\
500 "packuswb %%mm3, %%mm0 \n\t"\
501 OP(%%mm0, (%1),%%mm7, q)\
502 "add $48, %0 \n\t"\
503 "add %3, %1 \n\t"\
504 "decl %2 \n\t"\
505 " jnz 1b \n\t"\
506 : "+a"(tmp), "+c"(dst), "+g"(h)\
507 : "S"((x86_reg)dstStride)\
508 : "memory"\
509 );\
510 tmp += 8 - size*24;\
511 dst += 8 - size*dstStride;\
512 }while(w--);\
513 }\
514 \
515 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
516 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
517 }\
518 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
519 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
520 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
521 }\
522 \
523 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
524 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
525 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
526 src += 8*srcStride;\
527 dst += 8*dstStride;\
528 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
529 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
530 }\
531 \
532 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
533 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
534 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
535 src += 8*dstStride;\
536 dst += 8*dstStride;\
537 src2 += 8*src2Stride;\
538 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
539 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
540 }\
541 \
542 static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
543 put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
544 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
545 }\
546 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
547 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
548 }\
549 \
550 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
551 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
552 }\
553 \
554 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
555 {\
556 __asm__ volatile(\
557 "movq (%1), %%mm0 \n\t"\
558 "movq 24(%1), %%mm1 \n\t"\
559 "psraw $5, %%mm0 \n\t"\
560 "psraw $5, %%mm1 \n\t"\
561 "packuswb %%mm0, %%mm0 \n\t"\
562 "packuswb %%mm1, %%mm1 \n\t"\
563 PAVGB" (%0), %%mm0 \n\t"\
564 PAVGB" (%0,%3), %%mm1 \n\t"\
565 OP(%%mm0, (%2), %%mm4, d)\
566 OP(%%mm1, (%2,%4), %%mm5, d)\
567 "lea (%0,%3,2), %0 \n\t"\
568 "lea (%2,%4,2), %2 \n\t"\
569 "movq 48(%1), %%mm0 \n\t"\
570 "movq 72(%1), %%mm1 \n\t"\
571 "psraw $5, %%mm0 \n\t"\
572 "psraw $5, %%mm1 \n\t"\
573 "packuswb %%mm0, %%mm0 \n\t"\
574 "packuswb %%mm1, %%mm1 \n\t"\
575 PAVGB" (%0), %%mm0 \n\t"\
576 PAVGB" (%0,%3), %%mm1 \n\t"\
577 OP(%%mm0, (%2), %%mm4, d)\
578 OP(%%mm1, (%2,%4), %%mm5, d)\
579 :"+a"(src8), "+c"(src16), "+d"(dst)\
580 :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
581 :"memory");\
582 }\
583 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
584 {\
585 do{\
586 __asm__ volatile(\
587 "movq (%1), %%mm0 \n\t"\
588 "movq 8(%1), %%mm1 \n\t"\
589 "movq 48(%1), %%mm2 \n\t"\
590 "movq 8+48(%1), %%mm3 \n\t"\
591 "psraw $5, %%mm0 \n\t"\
592 "psraw $5, %%mm1 \n\t"\
593 "psraw $5, %%mm2 \n\t"\
594 "psraw $5, %%mm3 \n\t"\
595 "packuswb %%mm1, %%mm0 \n\t"\
596 "packuswb %%mm3, %%mm2 \n\t"\
597 PAVGB" (%0), %%mm0 \n\t"\
598 PAVGB" (%0,%3), %%mm2 \n\t"\
599 OP(%%mm0, (%2), %%mm5, q)\
600 OP(%%mm2, (%2,%4), %%mm5, q)\
601 ::"a"(src8), "c"(src16), "d"(dst),\
602 "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
603 :"memory");\
604 src8 += 2L*src8Stride;\
605 src16 += 48;\
606 dst += 2L*dstStride;\
607 }while(h-=2);\
608 }\
609 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
610 {\
611 OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
612 OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
613 }\
614
615
616 #if ARCH_X86_64
617 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
618 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
619 int h=16;\
620 __asm__ volatile(\
621 "pxor %%xmm15, %%xmm15 \n\t"\
622 "movdqa %6, %%xmm14 \n\t"\
623 "movdqa %7, %%xmm13 \n\t"\
624 "1: \n\t"\
625 "lddqu 6(%0), %%xmm1 \n\t"\
626 "lddqu -2(%0), %%xmm7 \n\t"\
627 "movdqa %%xmm1, %%xmm0 \n\t"\
628 "punpckhbw %%xmm15, %%xmm1 \n\t"\
629 "punpcklbw %%xmm15, %%xmm0 \n\t"\
630 "punpcklbw %%xmm15, %%xmm7 \n\t"\
631 "movdqa %%xmm1, %%xmm2 \n\t"\
632 "movdqa %%xmm0, %%xmm6 \n\t"\
633 "movdqa %%xmm1, %%xmm3 \n\t"\
634 "movdqa %%xmm0, %%xmm8 \n\t"\
635 "movdqa %%xmm1, %%xmm4 \n\t"\
636 "movdqa %%xmm0, %%xmm9 \n\t"\
637 "movdqa %%xmm0, %%xmm12 \n\t"\
638 "movdqa %%xmm1, %%xmm11 \n\t"\
639 "palignr $10,%%xmm0, %%xmm11\n\t"\
640 "palignr $10,%%xmm7, %%xmm12\n\t"\
641 "palignr $2, %%xmm0, %%xmm4 \n\t"\
642 "palignr $2, %%xmm7, %%xmm9 \n\t"\
643 "palignr $4, %%xmm0, %%xmm3 \n\t"\
644 "palignr $4, %%xmm7, %%xmm8 \n\t"\
645 "palignr $6, %%xmm0, %%xmm2 \n\t"\
646 "palignr $6, %%xmm7, %%xmm6 \n\t"\
647 "paddw %%xmm0 ,%%xmm11 \n\t"\
648 "palignr $8, %%xmm0, %%xmm1 \n\t"\
649 "palignr $8, %%xmm7, %%xmm0 \n\t"\
650 "paddw %%xmm12,%%xmm7 \n\t"\
651 "paddw %%xmm3, %%xmm2 \n\t"\
652 "paddw %%xmm8, %%xmm6 \n\t"\
653 "paddw %%xmm4, %%xmm1 \n\t"\
654 "paddw %%xmm9, %%xmm0 \n\t"\
655 "psllw $2, %%xmm2 \n\t"\
656 "psllw $2, %%xmm6 \n\t"\
657 "psubw %%xmm1, %%xmm2 \n\t"\
658 "psubw %%xmm0, %%xmm6 \n\t"\
659 "paddw %%xmm13,%%xmm11 \n\t"\
660 "paddw %%xmm13,%%xmm7 \n\t"\
661 "pmullw %%xmm14,%%xmm2 \n\t"\
662 "pmullw %%xmm14,%%xmm6 \n\t"\
663 "lddqu (%2), %%xmm3 \n\t"\
664 "paddw %%xmm11,%%xmm2 \n\t"\
665 "paddw %%xmm7, %%xmm6 \n\t"\
666 "psraw $5, %%xmm2 \n\t"\
667 "psraw $5, %%xmm6 \n\t"\
668 "packuswb %%xmm2,%%xmm6 \n\t"\
669 "pavgb %%xmm3, %%xmm6 \n\t"\
670 OP(%%xmm6, (%1), %%xmm4, dqa)\
671 "add %5, %0 \n\t"\
672 "add %5, %1 \n\t"\
673 "add %4, %2 \n\t"\
674 "decl %3 \n\t"\
675 "jg 1b \n\t"\
676 : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
677 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
678 "m"(ff_pw_5), "m"(ff_pw_16)\
679 : "memory"\
680 );\
681 }
682 #else // ARCH_X86_64
683 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
684 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
685 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
686 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
687 src += 8*dstStride;\
688 dst += 8*dstStride;\
689 src2 += 8*src2Stride;\
690 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
691 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
692 }
693 #endif // ARCH_X86_64
694
695 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
696 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
697 int h=8;\
698 __asm__ volatile(\
699 "pxor %%xmm7, %%xmm7 \n\t"\
700 "movdqa %0, %%xmm6 \n\t"\
701 :: "m"(ff_pw_5)\
702 );\
703 do{\
704 __asm__ volatile(\
705 "lddqu -2(%0), %%xmm1 \n\t"\
706 "movdqa %%xmm1, %%xmm0 \n\t"\
707 "punpckhbw %%xmm7, %%xmm1 \n\t"\
708 "punpcklbw %%xmm7, %%xmm0 \n\t"\
709 "movdqa %%xmm1, %%xmm2 \n\t"\
710 "movdqa %%xmm1, %%xmm3 \n\t"\
711 "movdqa %%xmm1, %%xmm4 \n\t"\
712 "movdqa %%xmm1, %%xmm5 \n\t"\
713 "palignr $2, %%xmm0, %%xmm4 \n\t"\
714 "palignr $4, %%xmm0, %%xmm3 \n\t"\
715 "palignr $6, %%xmm0, %%xmm2 \n\t"\
716 "palignr $8, %%xmm0, %%xmm1 \n\t"\
717 "palignr $10,%%xmm0, %%xmm5 \n\t"\
718 "paddw %%xmm5, %%xmm0 \n\t"\
719 "paddw %%xmm3, %%xmm2 \n\t"\
720 "paddw %%xmm4, %%xmm1 \n\t"\
721 "psllw $2, %%xmm2 \n\t"\
722 "movq (%2), %%xmm3 \n\t"\
723 "psubw %%xmm1, %%xmm2 \n\t"\
724 "paddw %5, %%xmm0 \n\t"\
725 "pmullw %%xmm6, %%xmm2 \n\t"\
726 "paddw %%xmm0, %%xmm2 \n\t"\
727 "psraw $5, %%xmm2 \n\t"\
728 "packuswb %%xmm2, %%xmm2 \n\t"\
729 "pavgb %%xmm3, %%xmm2 \n\t"\
730 OP(%%xmm2, (%1), %%xmm4, q)\
731 "add %4, %0 \n\t"\
732 "add %4, %1 \n\t"\
733 "add %3, %2 \n\t"\
734 : "+a"(src), "+c"(dst), "+d"(src2)\
735 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
736 "m"(ff_pw_16)\
737 : "memory"\
738 );\
739 }while(--h);\
740 }\
741 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
742 \
743 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
744 int h=8;\
745 __asm__ volatile(\
746 "pxor %%xmm7, %%xmm7 \n\t"\
747 "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
748 "1: \n\t"\
749 "lddqu -2(%0), %%xmm1 \n\t"\
750 "movdqa %%xmm1, %%xmm0 \n\t"\
751 "punpckhbw %%xmm7, %%xmm1 \n\t"\
752 "punpcklbw %%xmm7, %%xmm0 \n\t"\
753 "movdqa %%xmm1, %%xmm2 \n\t"\
754 "movdqa %%xmm1, %%xmm3 \n\t"\
755 "movdqa %%xmm1, %%xmm4 \n\t"\
756 "movdqa %%xmm1, %%xmm5 \n\t"\
757 "palignr $2, %%xmm0, %%xmm4 \n\t"\
758 "palignr $4, %%xmm0, %%xmm3 \n\t"\
759 "palignr $6, %%xmm0, %%xmm2 \n\t"\
760 "palignr $8, %%xmm0, %%xmm1 \n\t"\
761 "palignr $10,%%xmm0, %%xmm5 \n\t"\
762 "paddw %%xmm5, %%xmm0 \n\t"\
763 "paddw %%xmm3, %%xmm2 \n\t"\
764 "paddw %%xmm4, %%xmm1 \n\t"\
765 "psllw $2, %%xmm2 \n\t"\
766 "psubw %%xmm1, %%xmm2 \n\t"\
767 "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
768 "pmullw %%xmm6, %%xmm2 \n\t"\
769 "paddw %%xmm0, %%xmm2 \n\t"\
770 "psraw $5, %%xmm2 \n\t"\
771 "packuswb %%xmm2, %%xmm2 \n\t"\
772 OP(%%xmm2, (%1), %%xmm4, q)\
773 "add %3, %0 \n\t"\
774 "add %4, %1 \n\t"\
775 "decl %2 \n\t"\
776 " jnz 1b \n\t"\
777 : "+a"(src), "+c"(dst), "+g"(h)\
778 : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
779 : "memory"\
780 );\
781 }\
782 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
783 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
784 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
785 src += 8*srcStride;\
786 dst += 8*dstStride;\
787 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
788 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
789 }\
790
791 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
792 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
793 src -= 2*srcStride;\
794 \
795 __asm__ volatile(\
796 "pxor %%xmm7, %%xmm7 \n\t"\
797 "movq (%0), %%xmm0 \n\t"\
798 "add %2, %0 \n\t"\
799 "movq (%0), %%xmm1 \n\t"\
800 "add %2, %0 \n\t"\
801 "movq (%0), %%xmm2 \n\t"\
802 "add %2, %0 \n\t"\
803 "movq (%0), %%xmm3 \n\t"\
804 "add %2, %0 \n\t"\
805 "movq (%0), %%xmm4 \n\t"\
806 "add %2, %0 \n\t"\
807 "punpcklbw %%xmm7, %%xmm0 \n\t"\
808 "punpcklbw %%xmm7, %%xmm1 \n\t"\
809 "punpcklbw %%xmm7, %%xmm2 \n\t"\
810 "punpcklbw %%xmm7, %%xmm3 \n\t"\
811 "punpcklbw %%xmm7, %%xmm4 \n\t"\
812 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
813 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
814 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
815 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
816 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
817 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
818 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
819 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
820 \
821 : "+a"(src), "+c"(dst)\
822 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
823 : "memory"\
824 );\
825 if(h==16){\
826 __asm__ volatile(\
827 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
828 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
829 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
830 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
831 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
832 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
833 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
834 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
835 \
836 : "+a"(src), "+c"(dst)\
837 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
838 : "memory"\
839 );\
840 }\
841 }\
842 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
843 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
844 }\
845 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
846 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
847 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
848 }
849
850 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
851 int w = (size+8)>>3;
852 src -= 2*srcStride+2;
853 while(w--){
854 __asm__ volatile(
855 "pxor %%xmm7, %%xmm7 \n\t"
856 "movq (%0), %%xmm0 \n\t"
857 "add %2, %0 \n\t"
858 "movq (%0), %%xmm1 \n\t"
859 "add %2, %0 \n\t"
860 "movq (%0), %%xmm2 \n\t"
861 "add %2, %0 \n\t"
862 "movq (%0), %%xmm3 \n\t"
863 "add %2, %0 \n\t"
864 "movq (%0), %%xmm4 \n\t"
865 "add %2, %0 \n\t"
866 "punpcklbw %%xmm7, %%xmm0 \n\t"
867 "punpcklbw %%xmm7, %%xmm1 \n\t"
868 "punpcklbw %%xmm7, %%xmm2 \n\t"
869 "punpcklbw %%xmm7, %%xmm3 \n\t"
870 "punpcklbw %%xmm7, %%xmm4 \n\t"
871 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
872 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
873 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
874 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
875 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
876 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
877 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
878 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
879 : "+a"(src)
880 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
881 : "memory"
882 );
883 if(size==16){
884 __asm__ volatile(
885 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 8*48)
886 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 9*48)
887 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
888 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
889 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
890 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
891 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
892 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
893 : "+a"(src)
894 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
895 : "memory"
896 );
897 }
898 tmp += 8;
899 src += 8 - (size+5)*srcStride;
900 }
901 }
902
903 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
904 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
905 int h = size;\
906 if(size == 16){\
907 __asm__ volatile(\
908 "1: \n\t"\
909 "movdqa 32(%0), %%xmm4 \n\t"\
910 "movdqa 16(%0), %%xmm5 \n\t"\
911 "movdqa (%0), %%xmm7 \n\t"\
912 "movdqa %%xmm4, %%xmm3 \n\t"\
913 "movdqa %%xmm4, %%xmm2 \n\t"\
914 "movdqa %%xmm4, %%xmm1 \n\t"\
915 "movdqa %%xmm4, %%xmm0 \n\t"\
916 "palignr $10, %%xmm5, %%xmm0 \n\t"\
917 "palignr $8, %%xmm5, %%xmm1 \n\t"\
918 "palignr $6, %%xmm5, %%xmm2 \n\t"\
919 "palignr $4, %%xmm5, %%xmm3 \n\t"\
920 "palignr $2, %%xmm5, %%xmm4 \n\t"\
921 "paddw %%xmm5, %%xmm0 \n\t"\
922 "paddw %%xmm4, %%xmm1 \n\t"\
923 "paddw %%xmm3, %%xmm2 \n\t"\
924 "movdqa %%xmm5, %%xmm6 \n\t"\
925 "movdqa %%xmm5, %%xmm4 \n\t"\
926 "movdqa %%xmm5, %%xmm3 \n\t"\
927 "palignr $8, %%xmm7, %%xmm4 \n\t"\
928 "palignr $2, %%xmm7, %%xmm6 \n\t"\
929 "palignr $10, %%xmm7, %%xmm3 \n\t"\
930 "paddw %%xmm6, %%xmm4 \n\t"\
931 "movdqa %%xmm5, %%xmm6 \n\t"\
932 "palignr $6, %%xmm7, %%xmm5 \n\t"\
933 "palignr $4, %%xmm7, %%xmm6 \n\t"\
934 "paddw %%xmm7, %%xmm3 \n\t"\
935 "paddw %%xmm6, %%xmm5 \n\t"\
936 \
937 "psubw %%xmm1, %%xmm0 \n\t"\
938 "psubw %%xmm4, %%xmm3 \n\t"\
939 "psraw $2, %%xmm0 \n\t"\
940 "psraw $2, %%xmm3 \n\t"\
941 "psubw %%xmm1, %%xmm0 \n\t"\
942 "psubw %%xmm4, %%xmm3 \n\t"\
943 "paddw %%xmm2, %%xmm0 \n\t"\
944 "paddw %%xmm5, %%xmm3 \n\t"\
945 "psraw $2, %%xmm0 \n\t"\
946 "psraw $2, %%xmm3 \n\t"\
947 "paddw %%xmm2, %%xmm0 \n\t"\
948 "paddw %%xmm5, %%xmm3 \n\t"\
949 "psraw $6, %%xmm0 \n\t"\
950 "psraw $6, %%xmm3 \n\t"\
951 "packuswb %%xmm0, %%xmm3 \n\t"\
952 OP(%%xmm3, (%1), %%xmm7, dqa)\
953 "add $48, %0 \n\t"\
954 "add %3, %1 \n\t"\
955 "decl %2 \n\t"\
956 " jnz 1b \n\t"\
957 : "+a"(tmp), "+c"(dst), "+g"(h)\
958 : "S"((x86_reg)dstStride)\
959 : "memory"\
960 );\
961 }else{\
962 __asm__ volatile(\
963 "1: \n\t"\
964 "movdqa 16(%0), %%xmm1 \n\t"\
965 "movdqa (%0), %%xmm0 \n\t"\
966 "movdqa %%xmm1, %%xmm2 \n\t"\
967 "movdqa %%xmm1, %%xmm3 \n\t"\
968 "movdqa %%xmm1, %%xmm4 \n\t"\
969 "movdqa %%xmm1, %%xmm5 \n\t"\
970 "palignr $10, %%xmm0, %%xmm5 \n\t"\
971 "palignr $8, %%xmm0, %%xmm4 \n\t"\
972 "palignr $6, %%xmm0, %%xmm3 \n\t"\
973 "palignr $4, %%xmm0, %%xmm2 \n\t"\
974 "palignr $2, %%xmm0, %%xmm1 \n\t"\
975 "paddw %%xmm5, %%xmm0 \n\t"\
976 "paddw %%xmm4, %%xmm1 \n\t"\
977 "paddw %%xmm3, %%xmm2 \n\t"\
978 "psubw %%xmm1, %%xmm0 \n\t"\
979 "psraw $2, %%xmm0 \n\t"\
980 "psubw %%xmm1, %%xmm0 \n\t"\
981 "paddw %%xmm2, %%xmm0 \n\t"\
982 "psraw $2, %%xmm0 \n\t"\
983 "paddw %%xmm2, %%xmm0 \n\t"\
984 "psraw $6, %%xmm0 \n\t"\
985 "packuswb %%xmm0, %%xmm0 \n\t"\
986 OP(%%xmm0, (%1), %%xmm7, q)\
987 "add $48, %0 \n\t"\
988 "add %3, %1 \n\t"\
989 "decl %2 \n\t"\
990 " jnz 1b \n\t"\
991 : "+a"(tmp), "+c"(dst), "+g"(h)\
992 : "S"((x86_reg)dstStride)\
993 : "memory"\
994 );\
995 }\
996 }
997
998 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
999 static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
1000 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
1001 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
1002 }\
1003 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1004 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
1005 }\
1006 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1007 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1008 }\
1009
1010 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1011 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1012 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1013 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1014 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1015 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1016 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1017 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1018
1019 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1020 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1021 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1022 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1023 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1024 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1025 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1026 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1027
1028 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1029 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1030 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1031 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1032
1033 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1034 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1035 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1036 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1037
1038 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1039 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1040
1041 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1042 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1043 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1044 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1045 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1046
1047 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1048 put_pixels16_sse2(dst, src, stride, 16);
1049 }
1050 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1051 avg_pixels16_sse2(dst, src, stride, 16);
1052 }
1053 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1054 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1055
1056 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1057 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1058 OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1059 }\
1060
1061 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1062 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1063 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1064 }\
1065 \
1066 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1067 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1068 }\
1069 \
1070 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1071 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1072 }\
1073
1074 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1075 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1076 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1077 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1078 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1079 }\
1080 \
1081 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1082 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1083 }\
1084 \
1085 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1086 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1087 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1088 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1089 }\
1090
1091 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1092 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1093 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1094 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1095 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1096 }\
1097 \
1098 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1099 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1100 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1101 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1102 }\
1103 \
1104 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1105 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1106 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1107 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1108 }\
1109 \
1110 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1111 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1112 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1113 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1114 }\
1115 \
1116 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1117 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
1118 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1119 }\
1120 \
1121 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1122 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1123 uint8_t * const halfHV= temp;\
1124 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1125 assert(((int)temp & 7) == 0);\
1126 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1127 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1128 }\
1129 \
1130 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1131 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1132 uint8_t * const halfHV= temp;\
1133 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1134 assert(((int)temp & 7) == 0);\
1135 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1136 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1137 }\
1138 \
1139 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1140 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1141 uint8_t * const halfHV= temp;\
1142 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1143 assert(((int)temp & 7) == 0);\
1144 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1145 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1146 }\
1147 \
1148 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1149 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1150 uint8_t * const halfHV= temp;\
1151 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1152 assert(((int)temp & 7) == 0);\
1153 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1154 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1155 }\
1156
1157 #define H264_MC_4816(MMX)\
1158 H264_MC(put_, 4, MMX, 8)\
1159 H264_MC(put_, 8, MMX, 8)\
1160 H264_MC(put_, 16,MMX, 8)\
1161 H264_MC(avg_, 4, MMX, 8)\
1162 H264_MC(avg_, 8, MMX, 8)\
1163 H264_MC(avg_, 16,MMX, 8)\
1164
1165 #define H264_MC_816(QPEL, XMM)\
1166 QPEL(put_, 8, XMM, 16)\
1167 QPEL(put_, 16,XMM, 16)\
1168 QPEL(avg_, 8, XMM, 16)\
1169 QPEL(avg_, 16,XMM, 16)\
1170
1171
1172 #define AVG_3DNOW_OP(a,b,temp, size) \
1173 "mov" #size " " #b ", " #temp " \n\t"\
1174 "pavgusb " #temp ", " #a " \n\t"\
1175 "mov" #size " " #a ", " #b " \n\t"
1176 #define AVG_MMX2_OP(a,b,temp, size) \
1177 "mov" #size " " #b ", " #temp " \n\t"\
1178 "pavgb " #temp ", " #a " \n\t"\
1179 "mov" #size " " #a ", " #b " \n\t"
1180
1181 #define PAVGB "pavgusb"
1182 QPEL_H264(put_, PUT_OP, 3dnow)
1183 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1184 #undef PAVGB
1185 #define PAVGB "pavgb"
1186 QPEL_H264(put_, PUT_OP, mmx2)
1187 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
1188 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
1189 QPEL_H264_V_XMM(avg_, AVG_MMX2_OP, sse2)
1190 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
1191 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, sse2)
1192 #if HAVE_SSSE3
1193 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
1194 QPEL_H264_H_XMM(avg_, AVG_MMX2_OP, ssse3)
1195 QPEL_H264_HV2_XMM(put_, PUT_OP, ssse3)
1196 QPEL_H264_HV2_XMM(avg_, AVG_MMX2_OP, ssse3)
1197 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
1198 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, ssse3)
1199 #endif
1200 #undef PAVGB
1201
1202 H264_MC_4816(3dnow)
1203 H264_MC_4816(mmx2)
1204 H264_MC_816(H264_MC_V, sse2)
1205 H264_MC_816(H264_MC_HV, sse2)
1206 #if HAVE_SSSE3
1207 H264_MC_816(H264_MC_H, ssse3)
1208 H264_MC_816(H264_MC_HV, ssse3)
1209 #endif