h264_qpel_mmx.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
3  * Copyright (c) 2011 Daniel Kang
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "dsputil_mmx.h"
23 
24 /***********************************/
25 /* motion compensation */
26 
27 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
28  "mov"#q" "#C", "#T" \n\t"\
29  "mov"#d" (%0), "#F" \n\t"\
30  "paddw "#D", "#T" \n\t"\
31  "psllw $2, "#T" \n\t"\
32  "psubw "#B", "#T" \n\t"\
33  "psubw "#E", "#T" \n\t"\
34  "punpcklbw "#Z", "#F" \n\t"\
35  "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
36  "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
37  "add %2, %0 \n\t"\
38  "paddw "#F", "#A" \n\t"\
39  "paddw "#A", "#T" \n\t"\
40  "psraw $5, "#T" \n\t"\
41  "packuswb "#T", "#T" \n\t"\
42  OP(T, (%1), A, d)\
43  "add %3, %1 \n\t"
44 
45 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
46  "mov"#q" "#C", "#T" \n\t"\
47  "mov"#d" (%0), "#F" \n\t"\
48  "paddw "#D", "#T" \n\t"\
49  "psllw $2, "#T" \n\t"\
50  "paddw "MANGLE(ff_pw_16)", "#A"\n\t"\
51  "psubw "#B", "#T" \n\t"\
52  "psubw "#E", "#T" \n\t"\
53  "punpcklbw "#Z", "#F" \n\t"\
54  "pmullw "MANGLE(ff_pw_5)", "#T"\n\t"\
55  "paddw "#F", "#A" \n\t"\
56  "add %2, %0 \n\t"\
57  "paddw "#A", "#T" \n\t"\
58  "mov"#q" "#T", "#OF"(%1) \n\t"
59 
60 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
61 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
62 #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)
63 #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)
64 
65 
66 #define QPEL_H264(OPNAME, OP, MMX)\
67 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
68  int h=4;\
69 \
70  __asm__ volatile(\
71  "pxor %%mm7, %%mm7 \n\t"\
72  "movq "MANGLE(ff_pw_5) ", %%mm4\n\t"\
73  "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
74  "1: \n\t"\
75  "movd -1(%0), %%mm1 \n\t"\
76  "movd (%0), %%mm2 \n\t"\
77  "movd 1(%0), %%mm3 \n\t"\
78  "movd 2(%0), %%mm0 \n\t"\
79  "punpcklbw %%mm7, %%mm1 \n\t"\
80  "punpcklbw %%mm7, %%mm2 \n\t"\
81  "punpcklbw %%mm7, %%mm3 \n\t"\
82  "punpcklbw %%mm7, %%mm0 \n\t"\
83  "paddw %%mm0, %%mm1 \n\t"\
84  "paddw %%mm3, %%mm2 \n\t"\
85  "movd -2(%0), %%mm0 \n\t"\
86  "movd 3(%0), %%mm3 \n\t"\
87  "punpcklbw %%mm7, %%mm0 \n\t"\
88  "punpcklbw %%mm7, %%mm3 \n\t"\
89  "paddw %%mm3, %%mm0 \n\t"\
90  "psllw $2, %%mm2 \n\t"\
91  "psubw %%mm1, %%mm2 \n\t"\
92  "pmullw %%mm4, %%mm2 \n\t"\
93  "paddw %%mm5, %%mm0 \n\t"\
94  "paddw %%mm2, %%mm0 \n\t"\
95  "psraw $5, %%mm0 \n\t"\
96  "packuswb %%mm0, %%mm0 \n\t"\
97  OP(%%mm0, (%1),%%mm6, d)\
98  "add %3, %0 \n\t"\
99  "add %4, %1 \n\t"\
100  "decl %2 \n\t"\
101  " jnz 1b \n\t"\
102  : "+a"(src), "+c"(dst), "+g"(h)\
103  : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
104  : "memory"\
105  );\
106 }\
107 static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
108  int h=4;\
109  __asm__ volatile(\
110  "pxor %%mm7, %%mm7 \n\t"\
111  "movq %0, %%mm4 \n\t"\
112  "movq %1, %%mm5 \n\t"\
113  :: "m"(ff_pw_5), "m"(ff_pw_16)\
114  );\
115  do{\
116  __asm__ volatile(\
117  "movd -1(%0), %%mm1 \n\t"\
118  "movd (%0), %%mm2 \n\t"\
119  "movd 1(%0), %%mm3 \n\t"\
120  "movd 2(%0), %%mm0 \n\t"\
121  "punpcklbw %%mm7, %%mm1 \n\t"\
122  "punpcklbw %%mm7, %%mm2 \n\t"\
123  "punpcklbw %%mm7, %%mm3 \n\t"\
124  "punpcklbw %%mm7, %%mm0 \n\t"\
125  "paddw %%mm0, %%mm1 \n\t"\
126  "paddw %%mm3, %%mm2 \n\t"\
127  "movd -2(%0), %%mm0 \n\t"\
128  "movd 3(%0), %%mm3 \n\t"\
129  "punpcklbw %%mm7, %%mm0 \n\t"\
130  "punpcklbw %%mm7, %%mm3 \n\t"\
131  "paddw %%mm3, %%mm0 \n\t"\
132  "psllw $2, %%mm2 \n\t"\
133  "psubw %%mm1, %%mm2 \n\t"\
134  "pmullw %%mm4, %%mm2 \n\t"\
135  "paddw %%mm5, %%mm0 \n\t"\
136  "paddw %%mm2, %%mm0 \n\t"\
137  "movd (%2), %%mm3 \n\t"\
138  "psraw $5, %%mm0 \n\t"\
139  "packuswb %%mm0, %%mm0 \n\t"\
140  PAVGB" %%mm3, %%mm0 \n\t"\
141  OP(%%mm0, (%1),%%mm6, d)\
142  "add %4, %0 \n\t"\
143  "add %4, %1 \n\t"\
144  "add %3, %2 \n\t"\
145  : "+a"(src), "+c"(dst), "+d"(src2)\
146  : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
147  : "memory"\
148  );\
149  }while(--h);\
150 }\
151 static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
152  src -= 2*srcStride;\
153  __asm__ volatile(\
154  "pxor %%mm7, %%mm7 \n\t"\
155  "movd (%0), %%mm0 \n\t"\
156  "add %2, %0 \n\t"\
157  "movd (%0), %%mm1 \n\t"\
158  "add %2, %0 \n\t"\
159  "movd (%0), %%mm2 \n\t"\
160  "add %2, %0 \n\t"\
161  "movd (%0), %%mm3 \n\t"\
162  "add %2, %0 \n\t"\
163  "movd (%0), %%mm4 \n\t"\
164  "add %2, %0 \n\t"\
165  "punpcklbw %%mm7, %%mm0 \n\t"\
166  "punpcklbw %%mm7, %%mm1 \n\t"\
167  "punpcklbw %%mm7, %%mm2 \n\t"\
168  "punpcklbw %%mm7, %%mm3 \n\t"\
169  "punpcklbw %%mm7, %%mm4 \n\t"\
170  QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
171  QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
172  QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
173  QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
174  \
175  : "+a"(src), "+c"(dst)\
176  : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
177  : "memory"\
178  );\
179 }\
180 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){\
181  int h=4;\
182  int w=3;\
183  src -= 2*srcStride+2;\
184  while(w--){\
185  __asm__ volatile(\
186  "pxor %%mm7, %%mm7 \n\t"\
187  "movd (%0), %%mm0 \n\t"\
188  "add %2, %0 \n\t"\
189  "movd (%0), %%mm1 \n\t"\
190  "add %2, %0 \n\t"\
191  "movd (%0), %%mm2 \n\t"\
192  "add %2, %0 \n\t"\
193  "movd (%0), %%mm3 \n\t"\
194  "add %2, %0 \n\t"\
195  "movd (%0), %%mm4 \n\t"\
196  "add %2, %0 \n\t"\
197  "punpcklbw %%mm7, %%mm0 \n\t"\
198  "punpcklbw %%mm7, %%mm1 \n\t"\
199  "punpcklbw %%mm7, %%mm2 \n\t"\
200  "punpcklbw %%mm7, %%mm3 \n\t"\
201  "punpcklbw %%mm7, %%mm4 \n\t"\
202  QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\
203  QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\
204  QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\
205  QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\
206  \
207  : "+a"(src)\
208  : "c"(tmp), "S"((x86_reg)srcStride)\
209  : "memory"\
210  );\
211  tmp += 4;\
212  src += 4 - 9*srcStride;\
213  }\
214  tmp -= 3*4;\
215  __asm__ volatile(\
216  "1: \n\t"\
217  "movq (%0), %%mm0 \n\t"\
218  "paddw 10(%0), %%mm0 \n\t"\
219  "movq 2(%0), %%mm1 \n\t"\
220  "paddw 8(%0), %%mm1 \n\t"\
221  "movq 4(%0), %%mm2 \n\t"\
222  "paddw 6(%0), %%mm2 \n\t"\
223  "psubw %%mm1, %%mm0 \n\t"/*a-b (abccba)*/\
224  "psraw $2, %%mm0 \n\t"/*(a-b)/4 */\
225  "psubw %%mm1, %%mm0 \n\t"/*(a-b)/4-b */\
226  "paddsw %%mm2, %%mm0 \n\t"\
227  "psraw $2, %%mm0 \n\t"/*((a-b)/4-b+c)/4 */\
228  "paddw %%mm2, %%mm0 \n\t"/*(a-5*b+20*c)/16 */\
229  "psraw $6, %%mm0 \n\t"\
230  "packuswb %%mm0, %%mm0 \n\t"\
231  OP(%%mm0, (%1),%%mm7, d)\
232  "add $24, %0 \n\t"\
233  "add %3, %1 \n\t"\
234  "decl %2 \n\t"\
235  " jnz 1b \n\t"\
236  : "+a"(tmp), "+c"(dst), "+g"(h)\
237  : "S"((x86_reg)dstStride)\
238  : "memory"\
239  );\
240 }\
241 \
242 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
243  int h=8;\
244  __asm__ volatile(\
245  "pxor %%mm7, %%mm7 \n\t"\
246  "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
247  "1: \n\t"\
248  "movq (%0), %%mm0 \n\t"\
249  "movq 1(%0), %%mm2 \n\t"\
250  "movq %%mm0, %%mm1 \n\t"\
251  "movq %%mm2, %%mm3 \n\t"\
252  "punpcklbw %%mm7, %%mm0 \n\t"\
253  "punpckhbw %%mm7, %%mm1 \n\t"\
254  "punpcklbw %%mm7, %%mm2 \n\t"\
255  "punpckhbw %%mm7, %%mm3 \n\t"\
256  "paddw %%mm2, %%mm0 \n\t"\
257  "paddw %%mm3, %%mm1 \n\t"\
258  "psllw $2, %%mm0 \n\t"\
259  "psllw $2, %%mm1 \n\t"\
260  "movq -1(%0), %%mm2 \n\t"\
261  "movq 2(%0), %%mm4 \n\t"\
262  "movq %%mm2, %%mm3 \n\t"\
263  "movq %%mm4, %%mm5 \n\t"\
264  "punpcklbw %%mm7, %%mm2 \n\t"\
265  "punpckhbw %%mm7, %%mm3 \n\t"\
266  "punpcklbw %%mm7, %%mm4 \n\t"\
267  "punpckhbw %%mm7, %%mm5 \n\t"\
268  "paddw %%mm4, %%mm2 \n\t"\
269  "paddw %%mm3, %%mm5 \n\t"\
270  "psubw %%mm2, %%mm0 \n\t"\
271  "psubw %%mm5, %%mm1 \n\t"\
272  "pmullw %%mm6, %%mm0 \n\t"\
273  "pmullw %%mm6, %%mm1 \n\t"\
274  "movd -2(%0), %%mm2 \n\t"\
275  "movd 7(%0), %%mm5 \n\t"\
276  "punpcklbw %%mm7, %%mm2 \n\t"\
277  "punpcklbw %%mm7, %%mm5 \n\t"\
278  "paddw %%mm3, %%mm2 \n\t"\
279  "paddw %%mm5, %%mm4 \n\t"\
280  "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
281  "paddw %%mm5, %%mm2 \n\t"\
282  "paddw %%mm5, %%mm4 \n\t"\
283  "paddw %%mm2, %%mm0 \n\t"\
284  "paddw %%mm4, %%mm1 \n\t"\
285  "psraw $5, %%mm0 \n\t"\
286  "psraw $5, %%mm1 \n\t"\
287  "packuswb %%mm1, %%mm0 \n\t"\
288  OP(%%mm0, (%1),%%mm5, q)\
289  "add %3, %0 \n\t"\
290  "add %4, %1 \n\t"\
291  "decl %2 \n\t"\
292  " jnz 1b \n\t"\
293  : "+a"(src), "+c"(dst), "+g"(h)\
294  : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
295  : "memory"\
296  );\
297 }\
298 \
299 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
300  int h=8;\
301  __asm__ volatile(\
302  "pxor %%mm7, %%mm7 \n\t"\
303  "movq "MANGLE(ff_pw_5)", %%mm6\n\t"\
304  "1: \n\t"\
305  "movq (%0), %%mm0 \n\t"\
306  "movq 1(%0), %%mm2 \n\t"\
307  "movq %%mm0, %%mm1 \n\t"\
308  "movq %%mm2, %%mm3 \n\t"\
309  "punpcklbw %%mm7, %%mm0 \n\t"\
310  "punpckhbw %%mm7, %%mm1 \n\t"\
311  "punpcklbw %%mm7, %%mm2 \n\t"\
312  "punpckhbw %%mm7, %%mm3 \n\t"\
313  "paddw %%mm2, %%mm0 \n\t"\
314  "paddw %%mm3, %%mm1 \n\t"\
315  "psllw $2, %%mm0 \n\t"\
316  "psllw $2, %%mm1 \n\t"\
317  "movq -1(%0), %%mm2 \n\t"\
318  "movq 2(%0), %%mm4 \n\t"\
319  "movq %%mm2, %%mm3 \n\t"\
320  "movq %%mm4, %%mm5 \n\t"\
321  "punpcklbw %%mm7, %%mm2 \n\t"\
322  "punpckhbw %%mm7, %%mm3 \n\t"\
323  "punpcklbw %%mm7, %%mm4 \n\t"\
324  "punpckhbw %%mm7, %%mm5 \n\t"\
325  "paddw %%mm4, %%mm2 \n\t"\
326  "paddw %%mm3, %%mm5 \n\t"\
327  "psubw %%mm2, %%mm0 \n\t"\
328  "psubw %%mm5, %%mm1 \n\t"\
329  "pmullw %%mm6, %%mm0 \n\t"\
330  "pmullw %%mm6, %%mm1 \n\t"\
331  "movd -2(%0), %%mm2 \n\t"\
332  "movd 7(%0), %%mm5 \n\t"\
333  "punpcklbw %%mm7, %%mm2 \n\t"\
334  "punpcklbw %%mm7, %%mm5 \n\t"\
335  "paddw %%mm3, %%mm2 \n\t"\
336  "paddw %%mm5, %%mm4 \n\t"\
337  "movq "MANGLE(ff_pw_16)", %%mm5\n\t"\
338  "paddw %%mm5, %%mm2 \n\t"\
339  "paddw %%mm5, %%mm4 \n\t"\
340  "paddw %%mm2, %%mm0 \n\t"\
341  "paddw %%mm4, %%mm1 \n\t"\
342  "psraw $5, %%mm0 \n\t"\
343  "psraw $5, %%mm1 \n\t"\
344  "movq (%2), %%mm4 \n\t"\
345  "packuswb %%mm1, %%mm0 \n\t"\
346  PAVGB" %%mm4, %%mm0 \n\t"\
347  OP(%%mm0, (%1),%%mm5, q)\
348  "add %5, %0 \n\t"\
349  "add %5, %1 \n\t"\
350  "add %4, %2 \n\t"\
351  "decl %3 \n\t"\
352  "jg 1b \n\t"\
353  : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
354  : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
355  : "memory"\
356  );\
357 }\
358 \
359 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
360  int w= 2;\
361  src -= 2*srcStride;\
362  \
363  while(w--){\
364  __asm__ volatile(\
365  "pxor %%mm7, %%mm7 \n\t"\
366  "movd (%0), %%mm0 \n\t"\
367  "add %2, %0 \n\t"\
368  "movd (%0), %%mm1 \n\t"\
369  "add %2, %0 \n\t"\
370  "movd (%0), %%mm2 \n\t"\
371  "add %2, %0 \n\t"\
372  "movd (%0), %%mm3 \n\t"\
373  "add %2, %0 \n\t"\
374  "movd (%0), %%mm4 \n\t"\
375  "add %2, %0 \n\t"\
376  "punpcklbw %%mm7, %%mm0 \n\t"\
377  "punpcklbw %%mm7, %%mm1 \n\t"\
378  "punpcklbw %%mm7, %%mm2 \n\t"\
379  "punpcklbw %%mm7, %%mm3 \n\t"\
380  "punpcklbw %%mm7, %%mm4 \n\t"\
381  QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
382  QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
383  QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
384  QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
385  QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
386  QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
387  QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
388  QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
389  "cmpl $16, %4 \n\t"\
390  "jne 2f \n\t"\
391  QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
392  QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
393  QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
394  QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
395  QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
396  QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
397  QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
398  QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
399  "2: \n\t"\
400  \
401  : "+a"(src), "+c"(dst)\
402  : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
403  : "memory"\
404  );\
405  src += 4-(h+5)*srcStride;\
406  dst += 4-h*dstStride;\
407  }\
408 }\
409 static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
410  int w = (size+8)>>2;\
411  src -= 2*srcStride+2;\
412  while(w--){\
413  __asm__ volatile(\
414  "pxor %%mm7, %%mm7 \n\t"\
415  "movd (%0), %%mm0 \n\t"\
416  "add %2, %0 \n\t"\
417  "movd (%0), %%mm1 \n\t"\
418  "add %2, %0 \n\t"\
419  "movd (%0), %%mm2 \n\t"\
420  "add %2, %0 \n\t"\
421  "movd (%0), %%mm3 \n\t"\
422  "add %2, %0 \n\t"\
423  "movd (%0), %%mm4 \n\t"\
424  "add %2, %0 \n\t"\
425  "punpcklbw %%mm7, %%mm0 \n\t"\
426  "punpcklbw %%mm7, %%mm1 \n\t"\
427  "punpcklbw %%mm7, %%mm2 \n\t"\
428  "punpcklbw %%mm7, %%mm3 \n\t"\
429  "punpcklbw %%mm7, %%mm4 \n\t"\
430  QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\
431  QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\
432  QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\
433  QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\
434  QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\
435  QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\
436  QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\
437  QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\
438  "cmpl $16, %3 \n\t"\
439  "jne 2f \n\t"\
440  QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 8*48)\
441  QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 9*48)\
442  QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
443  QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
444  QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\
445  QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\
446  QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\
447  QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\
448  "2: \n\t"\
449  : "+a"(src)\
450  : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)\
451  : "memory"\
452  );\
453  tmp += 4;\
454  src += 4 - (size+5)*srcStride;\
455  }\
456 }\
457 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
458  int w = size>>4;\
459  do{\
460  int h = size;\
461  __asm__ volatile(\
462  "1: \n\t"\
463  "movq (%0), %%mm0 \n\t"\
464  "movq 8(%0), %%mm3 \n\t"\
465  "movq 2(%0), %%mm1 \n\t"\
466  "movq 10(%0), %%mm4 \n\t"\
467  "paddw %%mm4, %%mm0 \n\t"\
468  "paddw %%mm3, %%mm1 \n\t"\
469  "paddw 18(%0), %%mm3 \n\t"\
470  "paddw 16(%0), %%mm4 \n\t"\
471  "movq 4(%0), %%mm2 \n\t"\
472  "movq 12(%0), %%mm5 \n\t"\
473  "paddw 6(%0), %%mm2 \n\t"\
474  "paddw 14(%0), %%mm5 \n\t"\
475  "psubw %%mm1, %%mm0 \n\t"\
476  "psubw %%mm4, %%mm3 \n\t"\
477  "psraw $2, %%mm0 \n\t"\
478  "psraw $2, %%mm3 \n\t"\
479  "psubw %%mm1, %%mm0 \n\t"\
480  "psubw %%mm4, %%mm3 \n\t"\
481  "paddsw %%mm2, %%mm0 \n\t"\
482  "paddsw %%mm5, %%mm3 \n\t"\
483  "psraw $2, %%mm0 \n\t"\
484  "psraw $2, %%mm3 \n\t"\
485  "paddw %%mm2, %%mm0 \n\t"\
486  "paddw %%mm5, %%mm3 \n\t"\
487  "psraw $6, %%mm0 \n\t"\
488  "psraw $6, %%mm3 \n\t"\
489  "packuswb %%mm3, %%mm0 \n\t"\
490  OP(%%mm0, (%1),%%mm7, q)\
491  "add $48, %0 \n\t"\
492  "add %3, %1 \n\t"\
493  "decl %2 \n\t"\
494  " jnz 1b \n\t"\
495  : "+a"(tmp), "+c"(dst), "+g"(h)\
496  : "S"((x86_reg)dstStride)\
497  : "memory"\
498  );\
499  tmp += 8 - size*24;\
500  dst += 8 - size*dstStride;\
501  }while(w--);\
502 }\
503 \
504 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
505  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
506 }\
507 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
508  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
509  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
510 }\
511 \
512 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
513  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
514  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
515  src += 8*srcStride;\
516  dst += 8*dstStride;\
517  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
518  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
519 }\
520 \
521 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
522  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
523  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
524  src += 8*dstStride;\
525  dst += 8*dstStride;\
526  src2 += 8*src2Stride;\
527  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
528  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
529 }\
530 \
531 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){\
532  put_h264_qpel8or16_hv1_lowpass_ ## MMX(tmp, src, tmpStride, srcStride, size);\
533  OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
534 }\
535 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
536  OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 8);\
537 }\
538 \
539 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
540  OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst , tmp , src , dstStride, tmpStride, srcStride, 16);\
541 }\
542 \
543 static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
544 {\
545  __asm__ volatile(\
546  "movq (%1), %%mm0 \n\t"\
547  "movq 24(%1), %%mm1 \n\t"\
548  "psraw $5, %%mm0 \n\t"\
549  "psraw $5, %%mm1 \n\t"\
550  "packuswb %%mm0, %%mm0 \n\t"\
551  "packuswb %%mm1, %%mm1 \n\t"\
552  PAVGB" (%0), %%mm0 \n\t"\
553  PAVGB" (%0,%3), %%mm1 \n\t"\
554  OP(%%mm0, (%2), %%mm4, d)\
555  OP(%%mm1, (%2,%4), %%mm5, d)\
556  "lea (%0,%3,2), %0 \n\t"\
557  "lea (%2,%4,2), %2 \n\t"\
558  "movq 48(%1), %%mm0 \n\t"\
559  "movq 72(%1), %%mm1 \n\t"\
560  "psraw $5, %%mm0 \n\t"\
561  "psraw $5, %%mm1 \n\t"\
562  "packuswb %%mm0, %%mm0 \n\t"\
563  "packuswb %%mm1, %%mm1 \n\t"\
564  PAVGB" (%0), %%mm0 \n\t"\
565  PAVGB" (%0,%3), %%mm1 \n\t"\
566  OP(%%mm0, (%2), %%mm4, d)\
567  OP(%%mm1, (%2,%4), %%mm5, d)\
568  :"+a"(src8), "+c"(src16), "+d"(dst)\
569  :"S"((x86_reg)src8Stride), "D"((x86_reg)dstStride)\
570  :"memory");\
571 }\
572 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
573 {\
574  do{\
575  __asm__ volatile(\
576  "movq (%1), %%mm0 \n\t"\
577  "movq 8(%1), %%mm1 \n\t"\
578  "movq 48(%1), %%mm2 \n\t"\
579  "movq 8+48(%1), %%mm3 \n\t"\
580  "psraw $5, %%mm0 \n\t"\
581  "psraw $5, %%mm1 \n\t"\
582  "psraw $5, %%mm2 \n\t"\
583  "psraw $5, %%mm3 \n\t"\
584  "packuswb %%mm1, %%mm0 \n\t"\
585  "packuswb %%mm3, %%mm2 \n\t"\
586  PAVGB" (%0), %%mm0 \n\t"\
587  PAVGB" (%0,%3), %%mm2 \n\t"\
588  OP(%%mm0, (%2), %%mm5, q)\
589  OP(%%mm2, (%2,%4), %%mm5, q)\
590  ::"a"(src8), "c"(src16), "d"(dst),\
591  "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
592  :"memory");\
593  src8 += 2L*src8Stride;\
594  src16 += 48;\
595  dst += 2L*dstStride;\
596  }while(h-=2);\
597 }\
598 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
599 {\
600  OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
601  OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
602 }\
603 
604 
605 #if ARCH_X86_64
606 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
607 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
608  int h=16;\
609  __asm__ volatile(\
610  "pxor %%xmm15, %%xmm15 \n\t"\
611  "movdqa %6, %%xmm14 \n\t"\
612  "movdqa %7, %%xmm13 \n\t"\
613  "1: \n\t"\
614  "lddqu 6(%0), %%xmm1 \n\t"\
615  "lddqu -2(%0), %%xmm7 \n\t"\
616  "movdqa %%xmm1, %%xmm0 \n\t"\
617  "punpckhbw %%xmm15, %%xmm1 \n\t"\
618  "punpcklbw %%xmm15, %%xmm0 \n\t"\
619  "punpcklbw %%xmm15, %%xmm7 \n\t"\
620  "movdqa %%xmm1, %%xmm2 \n\t"\
621  "movdqa %%xmm0, %%xmm6 \n\t"\
622  "movdqa %%xmm1, %%xmm3 \n\t"\
623  "movdqa %%xmm0, %%xmm8 \n\t"\
624  "movdqa %%xmm1, %%xmm4 \n\t"\
625  "movdqa %%xmm0, %%xmm9 \n\t"\
626  "movdqa %%xmm0, %%xmm12 \n\t"\
627  "movdqa %%xmm1, %%xmm11 \n\t"\
628  "palignr $10,%%xmm0, %%xmm11\n\t"\
629  "palignr $10,%%xmm7, %%xmm12\n\t"\
630  "palignr $2, %%xmm0, %%xmm4 \n\t"\
631  "palignr $2, %%xmm7, %%xmm9 \n\t"\
632  "palignr $4, %%xmm0, %%xmm3 \n\t"\
633  "palignr $4, %%xmm7, %%xmm8 \n\t"\
634  "palignr $6, %%xmm0, %%xmm2 \n\t"\
635  "palignr $6, %%xmm7, %%xmm6 \n\t"\
636  "paddw %%xmm0 ,%%xmm11 \n\t"\
637  "palignr $8, %%xmm0, %%xmm1 \n\t"\
638  "palignr $8, %%xmm7, %%xmm0 \n\t"\
639  "paddw %%xmm12,%%xmm7 \n\t"\
640  "paddw %%xmm3, %%xmm2 \n\t"\
641  "paddw %%xmm8, %%xmm6 \n\t"\
642  "paddw %%xmm4, %%xmm1 \n\t"\
643  "paddw %%xmm9, %%xmm0 \n\t"\
644  "psllw $2, %%xmm2 \n\t"\
645  "psllw $2, %%xmm6 \n\t"\
646  "psubw %%xmm1, %%xmm2 \n\t"\
647  "psubw %%xmm0, %%xmm6 \n\t"\
648  "paddw %%xmm13,%%xmm11 \n\t"\
649  "paddw %%xmm13,%%xmm7 \n\t"\
650  "pmullw %%xmm14,%%xmm2 \n\t"\
651  "pmullw %%xmm14,%%xmm6 \n\t"\
652  "lddqu (%2), %%xmm3 \n\t"\
653  "paddw %%xmm11,%%xmm2 \n\t"\
654  "paddw %%xmm7, %%xmm6 \n\t"\
655  "psraw $5, %%xmm2 \n\t"\
656  "psraw $5, %%xmm6 \n\t"\
657  "packuswb %%xmm2,%%xmm6 \n\t"\
658  "pavgb %%xmm3, %%xmm6 \n\t"\
659  OP(%%xmm6, (%1), %%xmm4, dqa)\
660  "add %5, %0 \n\t"\
661  "add %5, %1 \n\t"\
662  "add %4, %2 \n\t"\
663  "decl %3 \n\t"\
664  "jg 1b \n\t"\
665  : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
666  : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
667  "m"(ff_pw_5), "m"(ff_pw_16)\
668  : XMM_CLOBBERS("%xmm0" , "%xmm1" , "%xmm2" , "%xmm3" , \
669  "%xmm4" , "%xmm5" , "%xmm6" , "%xmm7" , \
670  "%xmm8" , "%xmm9" , "%xmm10", "%xmm11", \
671  "%xmm12", "%xmm13", "%xmm14", "%xmm15",)\
672  "memory"\
673  );\
674 }
675 #else // ARCH_X86_64
676 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
677 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
678  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
679  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
680  src += 8*dstStride;\
681  dst += 8*dstStride;\
682  src2 += 8*src2Stride;\
683  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
684  OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
685 }
686 #endif // ARCH_X86_64
687 
688 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
689 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
690  int h=8;\
691  __asm__ volatile(\
692  "pxor %%xmm7, %%xmm7 \n\t"\
693  "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
694  "1: \n\t"\
695  "lddqu -2(%0), %%xmm1 \n\t"\
696  "movdqa %%xmm1, %%xmm0 \n\t"\
697  "punpckhbw %%xmm7, %%xmm1 \n\t"\
698  "punpcklbw %%xmm7, %%xmm0 \n\t"\
699  "movdqa %%xmm1, %%xmm2 \n\t"\
700  "movdqa %%xmm1, %%xmm3 \n\t"\
701  "movdqa %%xmm1, %%xmm4 \n\t"\
702  "movdqa %%xmm1, %%xmm5 \n\t"\
703  "palignr $2, %%xmm0, %%xmm4 \n\t"\
704  "palignr $4, %%xmm0, %%xmm3 \n\t"\
705  "palignr $6, %%xmm0, %%xmm2 \n\t"\
706  "palignr $8, %%xmm0, %%xmm1 \n\t"\
707  "palignr $10,%%xmm0, %%xmm5 \n\t"\
708  "paddw %%xmm5, %%xmm0 \n\t"\
709  "paddw %%xmm3, %%xmm2 \n\t"\
710  "paddw %%xmm4, %%xmm1 \n\t"\
711  "psllw $2, %%xmm2 \n\t"\
712  "movq (%2), %%xmm3 \n\t"\
713  "psubw %%xmm1, %%xmm2 \n\t"\
714  "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
715  "pmullw %%xmm6, %%xmm2 \n\t"\
716  "paddw %%xmm0, %%xmm2 \n\t"\
717  "psraw $5, %%xmm2 \n\t"\
718  "packuswb %%xmm2, %%xmm2 \n\t"\
719  "pavgb %%xmm3, %%xmm2 \n\t"\
720  OP(%%xmm2, (%1), %%xmm4, q)\
721  "add %5, %0 \n\t"\
722  "add %5, %1 \n\t"\
723  "add %4, %2 \n\t"\
724  "decl %3 \n\t"\
725  "jg 1b \n\t"\
726  : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
727  : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
728  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
729  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
730  "memory"\
731  );\
732 }\
733 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
734 \
735 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
736  int h=8;\
737  __asm__ volatile(\
738  "pxor %%xmm7, %%xmm7 \n\t"\
739  "movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
740  "1: \n\t"\
741  "lddqu -2(%0), %%xmm1 \n\t"\
742  "movdqa %%xmm1, %%xmm0 \n\t"\
743  "punpckhbw %%xmm7, %%xmm1 \n\t"\
744  "punpcklbw %%xmm7, %%xmm0 \n\t"\
745  "movdqa %%xmm1, %%xmm2 \n\t"\
746  "movdqa %%xmm1, %%xmm3 \n\t"\
747  "movdqa %%xmm1, %%xmm4 \n\t"\
748  "movdqa %%xmm1, %%xmm5 \n\t"\
749  "palignr $2, %%xmm0, %%xmm4 \n\t"\
750  "palignr $4, %%xmm0, %%xmm3 \n\t"\
751  "palignr $6, %%xmm0, %%xmm2 \n\t"\
752  "palignr $8, %%xmm0, %%xmm1 \n\t"\
753  "palignr $10,%%xmm0, %%xmm5 \n\t"\
754  "paddw %%xmm5, %%xmm0 \n\t"\
755  "paddw %%xmm3, %%xmm2 \n\t"\
756  "paddw %%xmm4, %%xmm1 \n\t"\
757  "psllw $2, %%xmm2 \n\t"\
758  "psubw %%xmm1, %%xmm2 \n\t"\
759  "paddw "MANGLE(ff_pw_16)", %%xmm0\n\t"\
760  "pmullw %%xmm6, %%xmm2 \n\t"\
761  "paddw %%xmm0, %%xmm2 \n\t"\
762  "psraw $5, %%xmm2 \n\t"\
763  "packuswb %%xmm2, %%xmm2 \n\t"\
764  OP(%%xmm2, (%1), %%xmm4, q)\
765  "add %3, %0 \n\t"\
766  "add %4, %1 \n\t"\
767  "decl %2 \n\t"\
768  " jnz 1b \n\t"\
769  : "+a"(src), "+c"(dst), "+g"(h)\
770  : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
771  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
772  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
773  "memory"\
774  );\
775 }\
776 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
777  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
778  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
779  src += 8*srcStride;\
780  dst += 8*dstStride;\
781  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
782  OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
783 }\
784 
785 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
786 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
787  src -= 2*srcStride;\
788  \
789  __asm__ volatile(\
790  "pxor %%xmm7, %%xmm7 \n\t"\
791  "movq (%0), %%xmm0 \n\t"\
792  "add %2, %0 \n\t"\
793  "movq (%0), %%xmm1 \n\t"\
794  "add %2, %0 \n\t"\
795  "movq (%0), %%xmm2 \n\t"\
796  "add %2, %0 \n\t"\
797  "movq (%0), %%xmm3 \n\t"\
798  "add %2, %0 \n\t"\
799  "movq (%0), %%xmm4 \n\t"\
800  "add %2, %0 \n\t"\
801  "punpcklbw %%xmm7, %%xmm0 \n\t"\
802  "punpcklbw %%xmm7, %%xmm1 \n\t"\
803  "punpcklbw %%xmm7, %%xmm2 \n\t"\
804  "punpcklbw %%xmm7, %%xmm3 \n\t"\
805  "punpcklbw %%xmm7, %%xmm4 \n\t"\
806  QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
807  QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
808  QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
809  QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
810  QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
811  QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
812  QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
813  QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
814  "cmpl $16, %4 \n\t"\
815  "jne 2f \n\t"\
816  QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
817  QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
818  QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
819  QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
820  QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
821  QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
822  QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
823  QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
824  "2: \n\t"\
825  \
826  : "+a"(src), "+c"(dst)\
827  : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "rm"(h)\
828  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
829  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
830  "memory"\
831  );\
832 }\
833 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
834  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
835 }\
836 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
837  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
838  OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
839 }
840 
841 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
842  int w = (size+8)>>3;
843  src -= 2*srcStride+2;
844  while(w--){
845  __asm__ volatile(
846  "pxor %%xmm7, %%xmm7 \n\t"
847  "movq (%0), %%xmm0 \n\t"
848  "add %2, %0 \n\t"
849  "movq (%0), %%xmm1 \n\t"
850  "add %2, %0 \n\t"
851  "movq (%0), %%xmm2 \n\t"
852  "add %2, %0 \n\t"
853  "movq (%0), %%xmm3 \n\t"
854  "add %2, %0 \n\t"
855  "movq (%0), %%xmm4 \n\t"
856  "add %2, %0 \n\t"
857  "punpcklbw %%xmm7, %%xmm0 \n\t"
858  "punpcklbw %%xmm7, %%xmm1 \n\t"
859  "punpcklbw %%xmm7, %%xmm2 \n\t"
860  "punpcklbw %%xmm7, %%xmm3 \n\t"
861  "punpcklbw %%xmm7, %%xmm4 \n\t"
862  QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
863  QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
864  QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
865  QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
866  QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
867  QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
868  QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
869  QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
870  "cmpl $16, %3 \n\t"
871  "jne 2f \n\t"
872  QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 8*48)
873  QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 9*48)
874  QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
875  QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
876  QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
877  QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
878  QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
879  QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
880  "2: \n\t"
881  : "+a"(src)
882  : "c"(tmp), "S"((x86_reg)srcStride), "rm"(size)
883  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",
884  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)
885  "memory"
886  );
887  tmp += 8;
888  src += 8 - (size+5)*srcStride;
889  }
890 }
891 
892 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
893 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
894  int h = size;\
895  if(size == 16){\
896  __asm__ volatile(\
897  "1: \n\t"\
898  "movdqa 32(%0), %%xmm4 \n\t"\
899  "movdqa 16(%0), %%xmm5 \n\t"\
900  "movdqa (%0), %%xmm7 \n\t"\
901  "movdqa %%xmm4, %%xmm3 \n\t"\
902  "movdqa %%xmm4, %%xmm2 \n\t"\
903  "movdqa %%xmm4, %%xmm1 \n\t"\
904  "movdqa %%xmm4, %%xmm0 \n\t"\
905  "palignr $10, %%xmm5, %%xmm0 \n\t"\
906  "palignr $8, %%xmm5, %%xmm1 \n\t"\
907  "palignr $6, %%xmm5, %%xmm2 \n\t"\
908  "palignr $4, %%xmm5, %%xmm3 \n\t"\
909  "palignr $2, %%xmm5, %%xmm4 \n\t"\
910  "paddw %%xmm5, %%xmm0 \n\t"\
911  "paddw %%xmm4, %%xmm1 \n\t"\
912  "paddw %%xmm3, %%xmm2 \n\t"\
913  "movdqa %%xmm5, %%xmm6 \n\t"\
914  "movdqa %%xmm5, %%xmm4 \n\t"\
915  "movdqa %%xmm5, %%xmm3 \n\t"\
916  "palignr $8, %%xmm7, %%xmm4 \n\t"\
917  "palignr $2, %%xmm7, %%xmm6 \n\t"\
918  "palignr $10, %%xmm7, %%xmm3 \n\t"\
919  "paddw %%xmm6, %%xmm4 \n\t"\
920  "movdqa %%xmm5, %%xmm6 \n\t"\
921  "palignr $6, %%xmm7, %%xmm5 \n\t"\
922  "palignr $4, %%xmm7, %%xmm6 \n\t"\
923  "paddw %%xmm7, %%xmm3 \n\t"\
924  "paddw %%xmm6, %%xmm5 \n\t"\
925  \
926  "psubw %%xmm1, %%xmm0 \n\t"\
927  "psubw %%xmm4, %%xmm3 \n\t"\
928  "psraw $2, %%xmm0 \n\t"\
929  "psraw $2, %%xmm3 \n\t"\
930  "psubw %%xmm1, %%xmm0 \n\t"\
931  "psubw %%xmm4, %%xmm3 \n\t"\
932  "paddw %%xmm2, %%xmm0 \n\t"\
933  "paddw %%xmm5, %%xmm3 \n\t"\
934  "psraw $2, %%xmm0 \n\t"\
935  "psraw $2, %%xmm3 \n\t"\
936  "paddw %%xmm2, %%xmm0 \n\t"\
937  "paddw %%xmm5, %%xmm3 \n\t"\
938  "psraw $6, %%xmm0 \n\t"\
939  "psraw $6, %%xmm3 \n\t"\
940  "packuswb %%xmm0, %%xmm3 \n\t"\
941  OP(%%xmm3, (%1), %%xmm7, dqa)\
942  "add $48, %0 \n\t"\
943  "add %3, %1 \n\t"\
944  "decl %2 \n\t"\
945  " jnz 1b \n\t"\
946  : "+a"(tmp), "+c"(dst), "+g"(h)\
947  : "S"((x86_reg)dstStride)\
948  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
949  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
950  "memory"\
951  );\
952  }else{\
953  __asm__ volatile(\
954  "1: \n\t"\
955  "movdqa 16(%0), %%xmm1 \n\t"\
956  "movdqa (%0), %%xmm0 \n\t"\
957  "movdqa %%xmm1, %%xmm2 \n\t"\
958  "movdqa %%xmm1, %%xmm3 \n\t"\
959  "movdqa %%xmm1, %%xmm4 \n\t"\
960  "movdqa %%xmm1, %%xmm5 \n\t"\
961  "palignr $10, %%xmm0, %%xmm5 \n\t"\
962  "palignr $8, %%xmm0, %%xmm4 \n\t"\
963  "palignr $6, %%xmm0, %%xmm3 \n\t"\
964  "palignr $4, %%xmm0, %%xmm2 \n\t"\
965  "palignr $2, %%xmm0, %%xmm1 \n\t"\
966  "paddw %%xmm5, %%xmm0 \n\t"\
967  "paddw %%xmm4, %%xmm1 \n\t"\
968  "paddw %%xmm3, %%xmm2 \n\t"\
969  "psubw %%xmm1, %%xmm0 \n\t"\
970  "psraw $2, %%xmm0 \n\t"\
971  "psubw %%xmm1, %%xmm0 \n\t"\
972  "paddw %%xmm2, %%xmm0 \n\t"\
973  "psraw $2, %%xmm0 \n\t"\
974  "paddw %%xmm2, %%xmm0 \n\t"\
975  "psraw $6, %%xmm0 \n\t"\
976  "packuswb %%xmm0, %%xmm0 \n\t"\
977  OP(%%xmm0, (%1), %%xmm7, q)\
978  "add $48, %0 \n\t"\
979  "add %3, %1 \n\t"\
980  "decl %2 \n\t"\
981  " jnz 1b \n\t"\
982  : "+a"(tmp), "+c"(dst), "+g"(h)\
983  : "S"((x86_reg)dstStride)\
984  : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
985  "%xmm4", "%xmm5", "%xmm6", "%xmm7",)\
986  "memory"\
987  );\
988  }\
989 }
990 
991 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
992 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){\
993  put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
994  OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
995 }\
996 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
997  OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
998 }\
999 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
1000  OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
1001 }\
1002 
1003 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
1004 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
1005 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
1006 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
1007 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
1008 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
1009 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
1010 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
1011 
1012 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
1013 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
1014 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
1015 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
1016 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
1017 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
1018 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
1019 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
1020 
1021 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
1022 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
1023 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
1024 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
1025 
1026 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
1027 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
1028 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
1029 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
1030 
1031 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
1032 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
1033 
1034 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
1035 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
1036 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
1037 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
1038 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
1039 
1040 static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1041  put_pixels16_sse2(dst, src, stride, 16);
1042 }
1043 static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
1044  avg_pixels16_sse2(dst, src, stride, 16);
1045 }
1046 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
1047 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
1048 
1049 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
1050 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
1051  OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
1052 }\
1053 
1054 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
1055 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1056  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
1057 }\
1058 \
1059 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1060  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
1061 }\
1062 \
1063 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1064  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
1065 }\
1066 
1067 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
1068 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1069  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1070  put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1071  OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
1072 }\
1073 \
1074 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1075  OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
1076 }\
1077 \
1078 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1079  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1080  put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1081  OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
1082 }\
1083 
1084 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
1085 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## 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 ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1089 }\
1090 \
1091 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1092  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1093  put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1094  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
1095 }\
1096 \
1097 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1098  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1099  put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
1100  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1101 }\
1102 \
1103 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1104  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
1105  put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
1106  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
1107 }\
1108 \
1109 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1110  DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
1111  OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
1112 }\
1113 \
1114 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1115  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1116  uint8_t * const halfHV= temp;\
1117  int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1118  assert(((int)temp & 7) == 0);\
1119  put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1120  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
1121 }\
1122 \
1123 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1124  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1125  uint8_t * const halfHV= temp;\
1126  int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1127  assert(((int)temp & 7) == 0);\
1128  put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1129  OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
1130 }\
1131 \
1132 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1133  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1134  uint8_t * const halfHV= temp;\
1135  int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1136  assert(((int)temp & 7) == 0);\
1137  put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1138  OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
1139 }\
1140 \
1141 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1142  DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
1143  uint8_t * const halfHV= temp;\
1144  int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
1145  assert(((int)temp & 7) == 0);\
1146  put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
1147  OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
1148 }\
1149 
1150 #define H264_MC_4816(MMX)\
1151 H264_MC(put_, 4, MMX, 8)\
1152 H264_MC(put_, 8, MMX, 8)\
1153 H264_MC(put_, 16,MMX, 8)\
1154 H264_MC(avg_, 4, MMX, 8)\
1155 H264_MC(avg_, 8, MMX, 8)\
1156 H264_MC(avg_, 16,MMX, 8)\
1157 
1158 #define H264_MC_816(QPEL, XMM)\
1159 QPEL(put_, 8, XMM, 16)\
1160 QPEL(put_, 16,XMM, 16)\
1161 QPEL(avg_, 8, XMM, 16)\
1162 QPEL(avg_, 16,XMM, 16)\
1163 
1164 
1165 #define AVG_3DNOW_OP(a,b,temp, size) \
1166 "mov" #size " " #b ", " #temp " \n\t"\
1167 "pavgusb " #temp ", " #a " \n\t"\
1168 "mov" #size " " #a ", " #b " \n\t"
1169 #define AVG_MMX2_OP(a,b,temp, size) \
1170 "mov" #size " " #b ", " #temp " \n\t"\
1171 "pavgb " #temp ", " #a " \n\t"\
1172 "mov" #size " " #a ", " #b " \n\t"
1173 
1174 #define PAVGB "pavgusb"
1175 QPEL_H264(put_, PUT_OP, 3dnow)
1176 QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
1177 #undef PAVGB
1178 #define PAVGB "pavgb"
1179 QPEL_H264(put_, PUT_OP, mmx2)
1180 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
1181 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
1182 QPEL_H264_V_XMM(avg_, AVG_MMX2_OP, sse2)
1183 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
1184 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, sse2)
1185 #if HAVE_SSSE3
1186 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
1187 QPEL_H264_H_XMM(avg_, AVG_MMX2_OP, ssse3)
1188 QPEL_H264_HV2_XMM(put_, PUT_OP, ssse3)
1189 QPEL_H264_HV2_XMM(avg_, AVG_MMX2_OP, ssse3)
1190 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
1191 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, ssse3)
1192 #endif
1193 #undef PAVGB
1194 
1195 H264_MC_4816(3dnow)
1196 H264_MC_4816(mmx2)
1197 H264_MC_816(H264_MC_V, sse2)
1198 H264_MC_816(H264_MC_HV, sse2)
1199 #if HAVE_SSSE3
1200 H264_MC_816(H264_MC_H, ssse3)
1201 H264_MC_816(H264_MC_HV, ssse3)
1202 #endif
1203 
1204 
1205 
1206 //10bit
1207 #define LUMA_MC_OP(OP, NUM, DEPTH, TYPE, OPT) \
1208 void ff_ ## OP ## _h264_qpel ## NUM ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT \
1209  (uint8_t *dst, uint8_t *src, int stride);
1210 
1211 #define LUMA_MC_ALL(DEPTH, TYPE, OPT) \
1212  LUMA_MC_OP(put, 4, DEPTH, TYPE, OPT) \
1213  LUMA_MC_OP(avg, 4, DEPTH, TYPE, OPT) \
1214  LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
1215  LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
1216  LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
1217  LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
1218 
1219 #define LUMA_MC_816(DEPTH, TYPE, OPT) \
1220  LUMA_MC_OP(put, 8, DEPTH, TYPE, OPT) \
1221  LUMA_MC_OP(avg, 8, DEPTH, TYPE, OPT) \
1222  LUMA_MC_OP(put, 16, DEPTH, TYPE, OPT) \
1223  LUMA_MC_OP(avg, 16, DEPTH, TYPE, OPT)
1224 
1225 LUMA_MC_ALL(10, mc00, mmxext)
1226 LUMA_MC_ALL(10, mc10, mmxext)
1227 LUMA_MC_ALL(10, mc20, mmxext)
1228 LUMA_MC_ALL(10, mc30, mmxext)
1229 LUMA_MC_ALL(10, mc01, mmxext)
1230 LUMA_MC_ALL(10, mc11, mmxext)
1231 LUMA_MC_ALL(10, mc21, mmxext)
1232 LUMA_MC_ALL(10, mc31, mmxext)
1233 LUMA_MC_ALL(10, mc02, mmxext)
1234 LUMA_MC_ALL(10, mc12, mmxext)
1235 LUMA_MC_ALL(10, mc22, mmxext)
1236 LUMA_MC_ALL(10, mc32, mmxext)
1237 LUMA_MC_ALL(10, mc03, mmxext)
1238 LUMA_MC_ALL(10, mc13, mmxext)
1239 LUMA_MC_ALL(10, mc23, mmxext)
1240 LUMA_MC_ALL(10, mc33, mmxext)
1241 
1242 LUMA_MC_816(10, mc00, sse2)
1243 LUMA_MC_816(10, mc10, sse2)
1244 LUMA_MC_816(10, mc10, sse2_cache64)
1245 LUMA_MC_816(10, mc10, ssse3_cache64)
1246 LUMA_MC_816(10, mc20, sse2)
1247 LUMA_MC_816(10, mc20, sse2_cache64)
1248 LUMA_MC_816(10, mc20, ssse3_cache64)
1249 LUMA_MC_816(10, mc30, sse2)
1250 LUMA_MC_816(10, mc30, sse2_cache64)
1251 LUMA_MC_816(10, mc30, ssse3_cache64)
1252 LUMA_MC_816(10, mc01, sse2)
1253 LUMA_MC_816(10, mc11, sse2)
1254 LUMA_MC_816(10, mc21, sse2)
1255 LUMA_MC_816(10, mc31, sse2)
1256 LUMA_MC_816(10, mc02, sse2)
1257 LUMA_MC_816(10, mc12, sse2)
1258 LUMA_MC_816(10, mc22, sse2)
1259 LUMA_MC_816(10, mc32, sse2)
1260 LUMA_MC_816(10, mc03, sse2)
1261 LUMA_MC_816(10, mc13, sse2)
1262 LUMA_MC_816(10, mc23, sse2)
1263 LUMA_MC_816(10, mc33, sse2)
1264 
1265 #define QPEL16_OPMC(OP, MC, MMX)\
1266 void ff_ ## OP ## _h264_qpel16_ ## MC ## _10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
1267  ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
1268  ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
1269  src += 8*stride;\
1270  dst += 8*stride;\
1271  ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst , src , stride);\
1272  ff_ ## OP ## _h264_qpel8_ ## MC ## _10_ ## MMX(dst+16, src+16, stride);\
1273 }
1274 
1275 #define QPEL16_OP(MC, MMX)\
1276 QPEL16_OPMC(put, MC, MMX)\
1277 QPEL16_OPMC(avg, MC, MMX)
1278 
1279 #define QPEL16(MMX)\
1280 QPEL16_OP(mc00, MMX)\
1281 QPEL16_OP(mc01, MMX)\
1282 QPEL16_OP(mc02, MMX)\
1283 QPEL16_OP(mc03, MMX)\
1284 QPEL16_OP(mc10, MMX)\
1285 QPEL16_OP(mc11, MMX)\
1286 QPEL16_OP(mc12, MMX)\
1287 QPEL16_OP(mc13, MMX)\
1288 QPEL16_OP(mc20, MMX)\
1289 QPEL16_OP(mc21, MMX)\
1290 QPEL16_OP(mc22, MMX)\
1291 QPEL16_OP(mc23, MMX)\
1292 QPEL16_OP(mc30, MMX)\
1293 QPEL16_OP(mc31, MMX)\
1294 QPEL16_OP(mc32, MMX)\
1295 QPEL16_OP(mc33, MMX)
1296 
1297 #if ARCH_X86_32 && HAVE_YASM // ARCH_X86_64 implies sse2+
1298 QPEL16(mmxext)
1299 #endif