FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vp9dsp.c
Go to the documentation of this file.
1 /*
2  * VP9 compatible video decoder
3  *
4  * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
5  * Copyright (C) 2013 Clément Bœsch <u pkh me>
6  *
7  * This file is part of FFmpeg.
8  *
9  * FFmpeg is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU Lesser General Public
11  * License as published by the Free Software Foundation; either
12  * version 2.1 of the License, or (at your option) any later version.
13  *
14  * FFmpeg is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17  * Lesser General Public License for more details.
18  *
19  * You should have received a copy of the GNU Lesser General Public
20  * License along with FFmpeg; if not, write to the Free Software
21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22  */
23 
24 #include "libavutil/common.h"
25 #include "libavutil/intreadwrite.h"
26 #include "vp9dsp.h"
27 #include "rnd_avg.h"
28 
29 // FIXME see whether we can merge parts of this (perhaps at least 4x4 and 8x8)
30 // back with h264pred.[ch]
31 
32 static void vert_4x4_c(uint8_t *dst, ptrdiff_t stride,
33  const uint8_t *left, const uint8_t *top)
34 {
35  unsigned p4 = AV_RN32A(top);
36 
37  AV_WN32A(dst + stride * 0, p4);
38  AV_WN32A(dst + stride * 1, p4);
39  AV_WN32A(dst + stride * 2, p4);
40  AV_WN32A(dst + stride * 3, p4);
41 }
42 
43 static void vert_8x8_c(uint8_t *dst, ptrdiff_t stride,
44  const uint8_t *left, const uint8_t *top)
45 {
46  uint64_t p8 = AV_RN64A(top);
47  int y;
48 
49  for (y = 0; y < 8; y++) {
50  AV_WN64A(dst, p8);
51  dst += stride;
52  }
53 }
54 
55 static void vert_16x16_c(uint8_t *dst, ptrdiff_t stride,
56  const uint8_t *left, const uint8_t *top)
57 {
58  uint64_t p8a = AV_RN64A(top + 0), p8b = AV_RN64A(top + 8);
59  int y;
60 
61  for (y = 0; y < 16; y++) {
62  AV_WN64A(dst + 0, p8a);
63  AV_WN64A(dst + 8, p8b);
64  dst += stride;
65  }
66 }
67 
68 static void vert_32x32_c(uint8_t *dst, ptrdiff_t stride,
69  const uint8_t *left, const uint8_t *top)
70 {
71  uint64_t p8a = AV_RN64A(top + 0), p8b = AV_RN64A(top + 8),
72  p8c = AV_RN64A(top + 16), p8d = AV_RN64A(top + 24);
73  int y;
74 
75  for (y = 0; y < 32; y++) {
76  AV_WN64A(dst + 0, p8a);
77  AV_WN64A(dst + 8, p8b);
78  AV_WN64A(dst + 16, p8c);
79  AV_WN64A(dst + 24, p8d);
80  dst += stride;
81  }
82 }
83 
84 static void hor_4x4_c(uint8_t *dst, ptrdiff_t stride,
85  const uint8_t *left, const uint8_t *top)
86 {
87  AV_WN32A(dst + stride * 0, left[3] * 0x01010101U);
88  AV_WN32A(dst + stride * 1, left[2] * 0x01010101U);
89  AV_WN32A(dst + stride * 2, left[1] * 0x01010101U);
90  AV_WN32A(dst + stride * 3, left[0] * 0x01010101U);
91 }
92 
93 static void hor_8x8_c(uint8_t *dst, ptrdiff_t stride,
94  const uint8_t *left, const uint8_t *top)
95 {
96  int y;
97 
98  for (y = 0; y < 8; y++) {
99  AV_WN64A(dst, left[7 - y] * 0x0101010101010101ULL);
100  dst += stride;
101  }
102 }
103 
104 static void hor_16x16_c(uint8_t *dst, ptrdiff_t stride,
105  const uint8_t *left, const uint8_t *top)
106 {
107  int y;
108 
109  for (y = 0; y < 16; y++) {
110  uint64_t p8 = left[15 - y] * 0x0101010101010101ULL;
111 
112  AV_WN64A(dst + 0, p8);
113  AV_WN64A(dst + 8, p8);
114  dst += stride;
115  }
116 }
117 
118 static void hor_32x32_c(uint8_t *dst, ptrdiff_t stride,
119  const uint8_t *left, const uint8_t *top)
120 {
121  int y;
122 
123  for (y = 0; y < 32; y++) {
124  uint64_t p8 = left[31 - y] * 0x0101010101010101ULL;
125 
126  AV_WN64A(dst + 0, p8);
127  AV_WN64A(dst + 8, p8);
128  AV_WN64A(dst + 16, p8);
129  AV_WN64A(dst + 24, p8);
130  dst += stride;
131  }
132 }
133 
134 static void tm_4x4_c(uint8_t *dst, ptrdiff_t stride,
135  const uint8_t *left, const uint8_t *top)
136 {
137  int y, tl = top[-1];
138 
139  for (y = 0; y < 4; y++) {
140  int l_m_tl = left[3 - y] - tl;
141 
142  dst[0] = av_clip_uint8(top[0] + l_m_tl);
143  dst[1] = av_clip_uint8(top[1] + l_m_tl);
144  dst[2] = av_clip_uint8(top[2] + l_m_tl);
145  dst[3] = av_clip_uint8(top[3] + l_m_tl);
146  dst += stride;
147  }
148 }
149 
150 static void tm_8x8_c(uint8_t *dst, ptrdiff_t stride,
151  const uint8_t *left, const uint8_t *top)
152 {
153  int y, tl = top[-1];
154 
155  for (y = 0; y < 8; y++) {
156  int l_m_tl = left[7 - y] - tl;
157 
158  dst[0] = av_clip_uint8(top[0] + l_m_tl);
159  dst[1] = av_clip_uint8(top[1] + l_m_tl);
160  dst[2] = av_clip_uint8(top[2] + l_m_tl);
161  dst[3] = av_clip_uint8(top[3] + l_m_tl);
162  dst[4] = av_clip_uint8(top[4] + l_m_tl);
163  dst[5] = av_clip_uint8(top[5] + l_m_tl);
164  dst[6] = av_clip_uint8(top[6] + l_m_tl);
165  dst[7] = av_clip_uint8(top[7] + l_m_tl);
166  dst += stride;
167  }
168 }
169 
170 static void tm_16x16_c(uint8_t *dst, ptrdiff_t stride,
171  const uint8_t *left, const uint8_t *top)
172 {
173  int y, tl = top[-1];
174 
175  for (y = 0; y < 16; y++) {
176  int l_m_tl = left[15 - y] - tl;
177 
178  dst[ 0] = av_clip_uint8(top[ 0] + l_m_tl);
179  dst[ 1] = av_clip_uint8(top[ 1] + l_m_tl);
180  dst[ 2] = av_clip_uint8(top[ 2] + l_m_tl);
181  dst[ 3] = av_clip_uint8(top[ 3] + l_m_tl);
182  dst[ 4] = av_clip_uint8(top[ 4] + l_m_tl);
183  dst[ 5] = av_clip_uint8(top[ 5] + l_m_tl);
184  dst[ 6] = av_clip_uint8(top[ 6] + l_m_tl);
185  dst[ 7] = av_clip_uint8(top[ 7] + l_m_tl);
186  dst[ 8] = av_clip_uint8(top[ 8] + l_m_tl);
187  dst[ 9] = av_clip_uint8(top[ 9] + l_m_tl);
188  dst[10] = av_clip_uint8(top[10] + l_m_tl);
189  dst[11] = av_clip_uint8(top[11] + l_m_tl);
190  dst[12] = av_clip_uint8(top[12] + l_m_tl);
191  dst[13] = av_clip_uint8(top[13] + l_m_tl);
192  dst[14] = av_clip_uint8(top[14] + l_m_tl);
193  dst[15] = av_clip_uint8(top[15] + l_m_tl);
194  dst += stride;
195  }
196 }
197 
198 static void tm_32x32_c(uint8_t *dst, ptrdiff_t stride,
199  const uint8_t *left, const uint8_t *top)
200 {
201  int y, tl = top[-1];
202 
203  for (y = 0; y < 32; y++) {
204  int l_m_tl = left[31 - y] - tl;
205 
206  dst[ 0] = av_clip_uint8(top[ 0] + l_m_tl);
207  dst[ 1] = av_clip_uint8(top[ 1] + l_m_tl);
208  dst[ 2] = av_clip_uint8(top[ 2] + l_m_tl);
209  dst[ 3] = av_clip_uint8(top[ 3] + l_m_tl);
210  dst[ 4] = av_clip_uint8(top[ 4] + l_m_tl);
211  dst[ 5] = av_clip_uint8(top[ 5] + l_m_tl);
212  dst[ 6] = av_clip_uint8(top[ 6] + l_m_tl);
213  dst[ 7] = av_clip_uint8(top[ 7] + l_m_tl);
214  dst[ 8] = av_clip_uint8(top[ 8] + l_m_tl);
215  dst[ 9] = av_clip_uint8(top[ 9] + l_m_tl);
216  dst[10] = av_clip_uint8(top[10] + l_m_tl);
217  dst[11] = av_clip_uint8(top[11] + l_m_tl);
218  dst[12] = av_clip_uint8(top[12] + l_m_tl);
219  dst[13] = av_clip_uint8(top[13] + l_m_tl);
220  dst[14] = av_clip_uint8(top[14] + l_m_tl);
221  dst[15] = av_clip_uint8(top[15] + l_m_tl);
222  dst[16] = av_clip_uint8(top[16] + l_m_tl);
223  dst[17] = av_clip_uint8(top[17] + l_m_tl);
224  dst[18] = av_clip_uint8(top[18] + l_m_tl);
225  dst[19] = av_clip_uint8(top[19] + l_m_tl);
226  dst[20] = av_clip_uint8(top[20] + l_m_tl);
227  dst[21] = av_clip_uint8(top[21] + l_m_tl);
228  dst[22] = av_clip_uint8(top[22] + l_m_tl);
229  dst[23] = av_clip_uint8(top[23] + l_m_tl);
230  dst[24] = av_clip_uint8(top[24] + l_m_tl);
231  dst[25] = av_clip_uint8(top[25] + l_m_tl);
232  dst[26] = av_clip_uint8(top[26] + l_m_tl);
233  dst[27] = av_clip_uint8(top[27] + l_m_tl);
234  dst[28] = av_clip_uint8(top[28] + l_m_tl);
235  dst[29] = av_clip_uint8(top[29] + l_m_tl);
236  dst[30] = av_clip_uint8(top[30] + l_m_tl);
237  dst[31] = av_clip_uint8(top[31] + l_m_tl);
238  dst += stride;
239  }
240 }
241 
242 static void dc_4x4_c(uint8_t *dst, ptrdiff_t stride,
243  const uint8_t *left, const uint8_t *top)
244 {
245  unsigned dc = 0x01010101U * ((left[0] + left[1] + left[2] + left[3] +
246  top[0] + top[1] + top[2] + top[3] + 4) >> 3);
247 
248  AV_WN32A(dst + stride * 0, dc);
249  AV_WN32A(dst + stride * 1, dc);
250  AV_WN32A(dst + stride * 2, dc);
251  AV_WN32A(dst + stride * 3, dc);
252 }
253 
254 static void dc_8x8_c(uint8_t *dst, ptrdiff_t stride,
255  const uint8_t *left, const uint8_t *top)
256 {
257  uint64_t dc = 0x0101010101010101ULL *
258  ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +
259  left[6] + left[7] + top[0] + top[1] + top[2] + top[3] +
260  top[4] + top[5] + top[6] + top[7] + 8) >> 4);
261  int y;
262 
263  for (y = 0; y < 8; y++) {
264  AV_WN64A(dst, dc);
265  dst += stride;
266  }
267 }
268 
269 static void dc_16x16_c(uint8_t *dst, ptrdiff_t stride,
270  const uint8_t *left, const uint8_t *top)
271 {
272  uint64_t dc = 0x0101010101010101ULL *
273  ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] + left[6] +
274  left[7] + left[8] + left[9] + left[10] + left[11] + left[12] +
275  left[13] + left[14] + left[15] + top[0] + top[1] + top[2] + top[3] +
276  top[4] + top[5] + top[6] + top[7] + top[8] + top[9] + top[10] +
277  top[11] + top[12] + top[13] + top[14] + top[15] + 16) >> 5);
278  int y;
279 
280  for (y = 0; y < 16; y++) {
281  AV_WN64A(dst + 0, dc);
282  AV_WN64A(dst + 8, dc);
283  dst += stride;
284  }
285 }
286 
287 static void dc_32x32_c(uint8_t *dst, ptrdiff_t stride,
288  const uint8_t *left, const uint8_t *top)
289 {
290  uint64_t dc = 0x0101010101010101ULL *
291  ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] + left[6] +
292  left[7] + left[8] + left[9] + left[10] + left[11] + left[12] +
293  left[13] + left[14] + left[15] + left[16] + left[17] + left[18] +
294  left[19] + left[20] + left[21] + left[22] + left[23] + left[24] +
295  left[25] + left[26] + left[27] + left[28] + left[29] + left[30] +
296  left[31] + top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +
297  top[6] + top[7] + top[8] + top[9] + top[10] + top[11] + top[12] +
298  top[13] + top[14] + top[15] + top[16] + top[17] + top[18] + top[19] +
299  top[20] + top[21] + top[22] + top[23] + top[24] + top[25] + top[26] +
300  top[27] + top[28] + top[29] + top[30] + top[31] + 32) >> 6);
301  int y;
302 
303  for (y = 0; y < 32; y++) {
304  AV_WN64A(dst + 0, dc);
305  AV_WN64A(dst + 8, dc);
306  AV_WN64A(dst + 16, dc);
307  AV_WN64A(dst + 24, dc);
308  dst += stride;
309  }
310 }
311 
312 static void dc_left_4x4_c(uint8_t *dst, ptrdiff_t stride,
313  const uint8_t *left, const uint8_t *top)
314 {
315  unsigned dc = 0x01010101U * ((left[0] + left[1] + left[2] + left[3] + 2) >> 2);
316 
317  AV_WN32A(dst + stride * 0, dc);
318  AV_WN32A(dst + stride * 1, dc);
319  AV_WN32A(dst + stride * 2, dc);
320  AV_WN32A(dst + stride * 3, dc);
321 }
322 
323 static void dc_left_8x8_c(uint8_t *dst, ptrdiff_t stride,
324  const uint8_t *left, const uint8_t *top)
325 {
326  uint64_t dc = 0x0101010101010101ULL *
327  ((left[0] + left[1] + left[2] + left[3] +
328  left[4] + left[5] + left[6] + left[7] + 4) >> 3);
329  int y;
330 
331  for (y = 0; y < 8; y++) {
332  AV_WN64A(dst, dc);
333  dst += stride;
334  }
335 }
336 
337 static void dc_left_16x16_c(uint8_t *dst, ptrdiff_t stride,
338  const uint8_t *left, const uint8_t *top)
339 {
340  uint64_t dc = 0x0101010101010101ULL *
341  ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +
342  left[6] + left[7] + left[8] + left[9] + left[10] + left[11] +
343  left[12] + left[13] + left[14] + left[15] + 8) >> 4);
344  int y;
345 
346  for (y = 0; y < 16; y++) {
347  AV_WN64A(dst + 0, dc);
348  AV_WN64A(dst + 8, dc);
349  dst += stride;
350  }
351 }
352 
353 static void dc_left_32x32_c(uint8_t *dst, ptrdiff_t stride,
354  const uint8_t *left, const uint8_t *top)
355 {
356  uint64_t dc = 0x0101010101010101ULL *
357  ((left[0] + left[1] + left[2] + left[3] + left[4] + left[5] +
358  left[6] + left[7] + left[8] + left[9] + left[10] + left[11] +
359  left[12] + left[13] + left[14] + left[15] + left[16] + left[17] +
360  left[18] + left[19] + left[20] + left[21] + left[22] + left[23] +
361  left[24] + left[25] + left[26] + left[27] + left[28] + left[29] +
362  left[30] + left[31] + 16) >> 5);
363  int y;
364 
365  for (y = 0; y < 32; y++) {
366  AV_WN64A(dst + 0, dc);
367  AV_WN64A(dst + 8, dc);
368  AV_WN64A(dst + 16, dc);
369  AV_WN64A(dst + 24, dc);
370  dst += stride;
371  }
372 }
373 
374 static void dc_top_4x4_c(uint8_t *dst, ptrdiff_t stride,
375  const uint8_t *left, const uint8_t *top)
376 {
377  unsigned dc = 0x01010101U * ((top[0] + top[1] + top[2] + top[3] + 2) >> 2);
378 
379  AV_WN32A(dst + stride * 0, dc);
380  AV_WN32A(dst + stride * 1, dc);
381  AV_WN32A(dst + stride * 2, dc);
382  AV_WN32A(dst + stride * 3, dc);
383 }
384 
385 static void dc_top_8x8_c(uint8_t *dst, ptrdiff_t stride,
386  const uint8_t *left, const uint8_t *top)
387 {
388  uint64_t dc = 0x0101010101010101ULL *
389  ((top[0] + top[1] + top[2] + top[3] +
390  top[4] + top[5] + top[6] + top[7] + 4) >> 3);
391  int y;
392 
393  for (y = 0; y < 8; y++) {
394  AV_WN64A(dst, dc);
395  dst += stride;
396  }
397 }
398 
399 static void dc_top_16x16_c(uint8_t *dst, ptrdiff_t stride,
400  const uint8_t *left, const uint8_t *top)
401 {
402  uint64_t dc = 0x0101010101010101ULL *
403  ((top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +
404  top[6] + top[7] + top[8] + top[9] + top[10] + top[11] +
405  top[12] + top[13] + top[14] + top[15] + 8) >> 4);
406  int y;
407 
408  for (y = 0; y < 16; y++) {
409  AV_WN64A(dst + 0, dc);
410  AV_WN64A(dst + 8, dc);
411  dst += stride;
412  }
413 }
414 
415 static void dc_top_32x32_c(uint8_t *dst, ptrdiff_t stride,
416  const uint8_t *left, const uint8_t *top)
417 {
418  uint64_t dc = 0x0101010101010101ULL *
419  ((top[0] + top[1] + top[2] + top[3] + top[4] + top[5] +
420  top[6] + top[7] + top[8] + top[9] + top[10] + top[11] +
421  top[12] + top[13] + top[14] + top[15] + top[16] + top[17] +
422  top[18] + top[19] + top[20] + top[21] + top[22] + top[23] +
423  top[24] + top[25] + top[26] + top[27] + top[28] + top[29] +
424  top[30] + top[31] + 16) >> 5);
425  int y;
426 
427  for (y = 0; y < 32; y++) {
428  AV_WN64A(dst + 0, dc);
429  AV_WN64A(dst + 8, dc);
430  AV_WN64A(dst + 16, dc);
431  AV_WN64A(dst + 24, dc);
432  dst += stride;
433  }
434 }
435 
436 static void dc_128_4x4_c(uint8_t *dst, ptrdiff_t stride,
437  const uint8_t *left, const uint8_t *top)
438 {
439  AV_WN32A(dst + stride * 0, 0x80808080U);
440  AV_WN32A(dst + stride * 1, 0x80808080U);
441  AV_WN32A(dst + stride * 2, 0x80808080U);
442  AV_WN32A(dst + stride * 3, 0x80808080U);
443 }
444 
445 static void dc_128_8x8_c(uint8_t *dst, ptrdiff_t stride,
446  const uint8_t *left, const uint8_t *top)
447 {
448  int y;
449 
450  for (y = 0; y < 8; y++) {
451  AV_WN64A(dst, 0x8080808080808080ULL);
452  dst += stride;
453  }
454 }
455 
456 static void dc_128_16x16_c(uint8_t *dst, ptrdiff_t stride,
457  const uint8_t *left, const uint8_t *top)
458 {
459  int y;
460 
461  for (y = 0; y < 16; y++) {
462  AV_WN64A(dst + 0, 0x8080808080808080ULL);
463  AV_WN64A(dst + 8, 0x8080808080808080ULL);
464  dst += stride;
465  }
466 }
467 
468 static void dc_128_32x32_c(uint8_t *dst, ptrdiff_t stride,
469  const uint8_t *left, const uint8_t *top)
470 {
471  int y;
472 
473  for (y = 0; y < 32; y++) {
474  AV_WN64A(dst + 0, 0x8080808080808080ULL);
475  AV_WN64A(dst + 8, 0x8080808080808080ULL);
476  AV_WN64A(dst + 16, 0x8080808080808080ULL);
477  AV_WN64A(dst + 24, 0x8080808080808080ULL);
478  dst += stride;
479  }
480 }
481 
482 static void dc_127_4x4_c(uint8_t *dst, ptrdiff_t stride,
483  const uint8_t *left, const uint8_t *top)
484 {
485  AV_WN32A(dst + stride * 0, 0x7F7F7F7FU);
486  AV_WN32A(dst + stride * 1, 0x7F7F7F7FU);
487  AV_WN32A(dst + stride * 2, 0x7F7F7F7FU);
488  AV_WN32A(dst + stride * 3, 0x7F7F7F7FU);
489 }
490 
491 static void dc_127_8x8_c(uint8_t *dst, ptrdiff_t stride,
492  const uint8_t *left, const uint8_t *top)
493 {
494  int y;
495 
496  for (y = 0; y < 8; y++) {
497  AV_WN64A(dst, 0x7F7F7F7F7F7F7F7FULL);
498  dst += stride;
499  }
500 }
501 
502 static void dc_127_16x16_c(uint8_t *dst, ptrdiff_t stride,
503  const uint8_t *left, const uint8_t *top)
504 {
505  int y;
506 
507  for (y = 0; y < 16; y++) {
508  AV_WN64A(dst + 0, 0x7F7F7F7F7F7F7F7FULL);
509  AV_WN64A(dst + 8, 0x7F7F7F7F7F7F7F7FULL);
510  dst += stride;
511  }
512 }
513 
514 static void dc_127_32x32_c(uint8_t *dst, ptrdiff_t stride,
515  const uint8_t *left, const uint8_t *top)
516 {
517  int y;
518 
519  for (y = 0; y < 32; y++) {
520  AV_WN64A(dst + 0, 0x7F7F7F7F7F7F7F7FULL);
521  AV_WN64A(dst + 8, 0x7F7F7F7F7F7F7F7FULL);
522  AV_WN64A(dst + 16, 0x7F7F7F7F7F7F7F7FULL);
523  AV_WN64A(dst + 24, 0x7F7F7F7F7F7F7F7FULL);
524  dst += stride;
525  }
526 }
527 
528 static void dc_129_4x4_c(uint8_t *dst, ptrdiff_t stride,
529  const uint8_t *left, const uint8_t *top)
530 {
531  AV_WN32A(dst + stride * 0, 0x81818181U);
532  AV_WN32A(dst + stride * 1, 0x81818181U);
533  AV_WN32A(dst + stride * 2, 0x81818181U);
534  AV_WN32A(dst + stride * 3, 0x81818181U);
535 }
536 
537 static void dc_129_8x8_c(uint8_t *dst, ptrdiff_t stride,
538  const uint8_t *left, const uint8_t *top)
539 {
540  int y;
541 
542  for (y = 0; y < 8; y++) {
543  AV_WN64A(dst, 0x8181818181818181ULL);
544  dst += stride;
545  }
546 }
547 
548 static void dc_129_16x16_c(uint8_t *dst, ptrdiff_t stride,
549  const uint8_t *left, const uint8_t *top)
550 {
551  int y;
552 
553  for (y = 0; y < 16; y++) {
554  AV_WN64A(dst + 0, 0x8181818181818181ULL);
555  AV_WN64A(dst + 8, 0x8181818181818181ULL);
556  dst += stride;
557  }
558 }
559 
560 static void dc_129_32x32_c(uint8_t *dst, ptrdiff_t stride,
561  const uint8_t *left, const uint8_t *top)
562 {
563  int y;
564 
565  for (y = 0; y < 32; y++) {
566  AV_WN64A(dst + 0, 0x8181818181818181ULL);
567  AV_WN64A(dst + 8, 0x8181818181818181ULL);
568  AV_WN64A(dst + 16, 0x8181818181818181ULL);
569  AV_WN64A(dst + 24, 0x8181818181818181ULL);
570  dst += stride;
571  }
572 }
573 
574 #define DST(x, y) dst[(x) + (y) * stride]
575 
576 static void diag_downleft_4x4_c(uint8_t *dst, ptrdiff_t stride,
577  const uint8_t *left, const uint8_t *top)
578 {
579  int a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
580  a4 = top[4], a5 = top[5], a6 = top[6], a7 = top[7];
581 
582  DST(0,0) = (a0 + a1 * 2 + a2 + 2) >> 2;
583  DST(1,0) = DST(0,1) = (a1 + a2 * 2 + a3 + 2) >> 2;
584  DST(2,0) = DST(1,1) = DST(0,2) = (a2 + a3 * 2 + a4 + 2) >> 2;
585  DST(3,0) = DST(2,1) = DST(1,2) = DST(0,3) = (a3 + a4 * 2 + a5 + 2) >> 2;
586  DST(3,1) = DST(2,2) = DST(1,3) = (a4 + a5 * 2 + a6 + 2) >> 2;
587  DST(3,2) = DST(2,3) = (a5 + a6 * 2 + a7 + 2) >> 2;
588  DST(3,3) = a7; // note: this is different from vp8 and such
589 }
590 
591 #define def_diag_downleft(size) \
592 static void diag_downleft_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
593  const uint8_t *left, const uint8_t *top) \
594 { \
595  int i, j; \
596  uint8_t v[size - 1]; \
597 \
598  for (i = 0; i < size - 2; i++) \
599  v[i] = (top[i] + top[i + 1] * 2 + top[i + 2] + 2) >> 2; \
600  v[size - 2] = (top[size - 2] + top[size - 1] * 3 + 2) >> 2; \
601 \
602  for (j = 0; j < size; j++) { \
603  memcpy(dst + j*stride, v + j, size - 1 - j); \
604  memset(dst + j*stride + size - 1 - j, top[size - 1], j + 1); \
605  } \
606 }
607 
611 
612 static void diag_downright_4x4_c(uint8_t *dst, ptrdiff_t stride,
613  const uint8_t *left, const uint8_t *top)
614 {
615  int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
616  l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0];
617 
618  DST(0,3) = (l1 + l2 * 2 + l3 + 2) >> 2;
619  DST(0,2) = DST(1,3) = (l0 + l1 * 2 + l2 + 2) >> 2;
620  DST(0,1) = DST(1,2) = DST(2,3) = (tl + l0 * 2 + l1 + 2) >> 2;
621  DST(0,0) = DST(1,1) = DST(2,2) = DST(3,3) = (l0 + tl * 2 + a0 + 2) >> 2;
622  DST(1,0) = DST(2,1) = DST(3,2) = (tl + a0 * 2 + a1 + 2) >> 2;
623  DST(2,0) = DST(3,1) = (a0 + a1 * 2 + a2 + 2) >> 2;
624  DST(3,0) = (a1 + a2 * 2 + a3 + 2) >> 2;
625 }
626 
627 #define def_diag_downright(size) \
628 static void diag_downright_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
629  const uint8_t *left, const uint8_t *top) \
630 { \
631  int i, j; \
632  uint8_t v[size + size - 1]; \
633 \
634  for (i = 0; i < size - 2; i++) { \
635  v[i ] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \
636  v[size + 1 + i] = (top[i] + top[i + 1] * 2 + top[i + 2] + 2) >> 2; \
637  } \
638  v[size - 2] = (left[size - 2] + left[size - 1] * 2 + top[-1] + 2) >> 2; \
639  v[size - 1] = (left[size - 1] + top[-1] * 2 + top[ 0] + 2) >> 2; \
640  v[size ] = (top[-1] + top[0] * 2 + top[ 1] + 2) >> 2; \
641 \
642  for (j = 0; j < size; j++) \
643  memcpy(dst + j*stride, v + size - 1 - j, size); \
644 }
645 
649 
650 static void vert_right_4x4_c(uint8_t *dst, ptrdiff_t stride,
651  const uint8_t *left, const uint8_t *top)
652 {
653  int tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
654  l0 = left[3], l1 = left[2], l2 = left[1];
655 
656  DST(0,3) = (l0 + l1 * 2 + l2 + 2) >> 2;
657  DST(0,2) = (tl + l0 * 2 + l1 + 2) >> 2;
658  DST(0,0) = DST(1,2) = (tl + a0 + 1) >> 1;
659  DST(0,1) = DST(1,3) = (l0 + tl * 2 + a0 + 2) >> 2;
660  DST(1,0) = DST(2,2) = (a0 + a1 + 1) >> 1;
661  DST(1,1) = DST(2,3) = (tl + a0 * 2 + a1 + 2) >> 2;
662  DST(2,0) = DST(3,2) = (a1 + a2 + 1) >> 1;
663  DST(2,1) = DST(3,3) = (a0 + a1 * 2 + a2 + 2) >> 2;
664  DST(3,0) = (a2 + a3 + 1) >> 1;
665  DST(3,1) = (a1 + a2 * 2 + a3 + 2) >> 2;
666 }
667 
668 #define def_vert_right(size) \
669 static void vert_right_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
670  const uint8_t *left, const uint8_t *top) \
671 { \
672  int i, j; \
673  uint8_t ve[size + size/2 - 1], vo[size + size/2 - 1]; \
674 \
675  for (i = 0; i < size/2 - 2; i++) { \
676  vo[i] = (left[i*2 + 3] + left[i*2 + 2] * 2 + left[i*2 + 1] + 2) >> 2; \
677  ve[i] = (left[i*2 + 4] + left[i*2 + 3] * 2 + left[i*2 + 2] + 2) >> 2; \
678  } \
679  vo[size/2 - 2] = (left[size - 1] + left[size - 2] * 2 + left[size - 3] + 2) >> 2; \
680  ve[size/2 - 2] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \
681 \
682  ve[size/2 - 1] = (top[-1] + top[0] + 1) >> 1; \
683  vo[size/2 - 1] = (left[size - 1] + top[-1] * 2 + top[0] + 2) >> 2; \
684  for (i = 0; i < size - 1; i++) { \
685  ve[size/2 + i] = (top[i] + top[i + 1] + 1) >> 1; \
686  vo[size/2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \
687  } \
688 \
689  for (j = 0; j < size / 2; j++) { \
690  memcpy(dst + j*2 *stride, ve + size/2 - 1 - j, size); \
691  memcpy(dst + (j*2 + 1)*stride, vo + size/2 - 1 - j, size); \
692  } \
693 }
694 
696 def_vert_right(16)
697 def_vert_right(32)
698 
699 static void hor_down_4x4_c(uint8_t *dst, ptrdiff_t stride,
700  const uint8_t *left, const uint8_t *top)
701 {
702  int l0 = left[3], l1 = left[2], l2 = left[1], l3 = left[0],
703  tl = top[-1], a0 = top[0], a1 = top[1], a2 = top[2];
704 
705  DST(2,0) = (tl + a0 * 2 + a1 + 2) >> 2;
706  DST(3,0) = (a0 + a1 * 2 + a2 + 2) >> 2;
707  DST(0,0) = DST(2,1) = (tl + l0 + 1) >> 1;
708  DST(1,0) = DST(3,1) = (a0 + tl * 2 + l0 + 2) >> 2;
709  DST(0,1) = DST(2,2) = (l0 + l1 + 1) >> 1;
710  DST(1,1) = DST(3,2) = (tl + l0 * 2 + l1 + 2) >> 2;
711  DST(0,2) = DST(2,3) = (l1 + l2 + 1) >> 1;
712  DST(1,2) = DST(3,3) = (l0 + l1 * 2 + l2 + 2) >> 2;
713  DST(0,3) = (l2 + l3 + 1) >> 1;
714  DST(1,3) = (l1 + l2 * 2 + l3 + 2) >> 2;
715 }
716 
717 #define def_hor_down(size) \
718 static void hor_down_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
719  const uint8_t *left, const uint8_t *top) \
720 { \
721  int i, j; \
722  uint8_t v[size * 3 - 2]; \
723 \
724  for (i = 0; i < size - 2; i++) { \
725  v[i*2 ] = (left[i + 1] + left[i + 0] + 1) >> 1; \
726  v[i*2 + 1] = (left[i + 2] + left[i + 1] * 2 + left[i + 0] + 2) >> 2; \
727  v[size*2 + i] = (top[i - 1] + top[i] * 2 + top[i + 1] + 2) >> 2; \
728  } \
729  v[size*2 - 2] = (top[-1] + left[size - 1] + 1) >> 1; \
730  v[size*2 - 4] = (left[size - 1] + left[size - 2] + 1) >> 1; \
731  v[size*2 - 1] = (top[0] + top[-1] * 2 + left[size - 1] + 2) >> 2; \
732  v[size*2 - 3] = (top[-1] + left[size - 1] * 2 + left[size - 2] + 2) >> 2; \
733 \
734  for (j = 0; j < size; j++) \
735  memcpy(dst + j*stride, v + size*2 - 2 - j*2, size); \
736 }
737 
739 def_hor_down(16)
740 def_hor_down(32)
741 
742 static void vert_left_4x4_c(uint8_t *dst, ptrdiff_t stride,
743  const uint8_t *left, const uint8_t *top)
744 {
745  int a0 = top[0], a1 = top[1], a2 = top[2], a3 = top[3],
746  a4 = top[4], a5 = top[5], a6 = top[6];
747 
748  DST(0,0) = (a0 + a1 + 1) >> 1;
749  DST(0,1) = (a0 + a1 * 2 + a2 + 2) >> 2;
750  DST(1,0) = DST(0,2) = (a1 + a2 + 1) >> 1;
751  DST(1,1) = DST(0,3) = (a1 + a2 * 2 + a3 + 2) >> 2;
752  DST(2,0) = DST(1,2) = (a2 + a3 + 1) >> 1;
753  DST(2,1) = DST(1,3) = (a2 + a3 * 2 + a4 + 2) >> 2;
754  DST(3,0) = DST(2,2) = (a3 + a4 + 1) >> 1;
755  DST(3,1) = DST(2,3) = (a3 + a4 * 2 + a5 + 2) >> 2;
756  DST(3,2) = (a4 + a5 + 1) >> 1;
757  DST(3,3) = (a4 + a5 * 2 + a6 + 2) >> 2;
758 }
759 
760 #define def_vert_left(size) \
761 static void vert_left_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
762  const uint8_t *left, const uint8_t *top) \
763 { \
764  int i, j; \
765  uint8_t ve[size - 1], vo[size - 1]; \
766 \
767  for (i = 0; i < size - 2; i++) { \
768  ve[i] = (top[i] + top[i + 1] + 1) >> 1; \
769  vo[i] = (top[i] + top[i + 1] * 2 + top[i + 2] + 2) >> 2; \
770  } \
771  ve[size - 2] = (top[size - 2] + top[size - 1] + 1) >> 1; \
772  vo[size - 2] = (top[size - 2] + top[size - 1] * 3 + 2) >> 2; \
773 \
774  for (j = 0; j < size / 2; j++) { \
775  memcpy(dst + j*2 * stride, ve + j, size - j - 1); \
776  memset(dst + j*2 * stride + size - j - 1, top[size - 1], j + 1); \
777  memcpy(dst + (j*2 + 1) * stride, vo + j, size - j - 1); \
778  memset(dst + (j*2 + 1) * stride + size - j - 1, top[size - 1], j + 1); \
779  } \
780 }
781 
783 def_vert_left(16)
784 def_vert_left(32)
785 
786 static void hor_up_4x4_c(uint8_t *dst, ptrdiff_t stride,
787  const uint8_t *left, const uint8_t *top)
788 {
789  int l0 = left[0], l1 = left[1], l2 = left[2], l3 = left[3];
790 
791  DST(0,0) = (l0 + l1 + 1) >> 1;
792  DST(1,0) = (l0 + l1 * 2 + l2 + 2) >> 2;
793  DST(0,1) = DST(2,0) = (l1 + l2 + 1) >> 1;
794  DST(1,1) = DST(3,0) = (l1 + l2 * 2 + l3 + 2) >> 2;
795  DST(0,2) = DST(2,1) = (l2 + l3 + 1) >> 1;
796  DST(1,2) = DST(3,1) = (l2 + l3 * 3 + 2) >> 2;
797  DST(0,3) = DST(1,3) = DST(2,2) = DST(2,3) = DST(3,2) = DST(3,3) = l3;
798 }
799 
800 #define def_hor_up(size) \
801 static void hor_up_##size##x##size##_c(uint8_t *dst, ptrdiff_t stride, \
802  const uint8_t *left, const uint8_t *top) \
803 { \
804  int i, j; \
805  uint8_t v[size*2 - 2]; \
806 \
807  for (i = 0; i < size - 2; i++) { \
808  v[i*2 ] = (left[i] + left[i + 1] + 1) >> 1; \
809  v[i*2 + 1] = (left[i] + left[i + 1] * 2 + left[i + 2] + 2) >> 2; \
810  } \
811  v[size*2 - 4] = (left[size - 2] + left[size - 1] + 1) >> 1; \
812  v[size*2 - 3] = (left[size - 2] + left[size - 1] * 3 + 2) >> 2; \
813 \
814  for (j = 0; j < size / 2; j++) \
815  memcpy(dst + j*stride, v + j*2, size); \
816  for (j = size / 2; j < size; j++) { \
817  memcpy(dst + j*stride, v + j*2, size*2 - 2 - j*2); \
818  memset(dst + j*stride + size*2 - 2 - j*2, left[size - 1], \
819  2 + j*2 - size); \
820  } \
821 }
822 
824 def_hor_up(16)
825 def_hor_up(32)
826 
827 #undef DST
828 
829 static av_cold void vp9dsp_intrapred_init(VP9DSPContext *dsp)
830 {
831 #define init_intra_pred(tx, sz) \
832  dsp->intra_pred[tx][VERT_PRED] = vert_##sz##_c; \
833  dsp->intra_pred[tx][HOR_PRED] = hor_##sz##_c; \
834  dsp->intra_pred[tx][DC_PRED] = dc_##sz##_c; \
835  dsp->intra_pred[tx][DIAG_DOWN_LEFT_PRED] = diag_downleft_##sz##_c; \
836  dsp->intra_pred[tx][DIAG_DOWN_RIGHT_PRED] = diag_downright_##sz##_c; \
837  dsp->intra_pred[tx][VERT_RIGHT_PRED] = vert_right_##sz##_c; \
838  dsp->intra_pred[tx][HOR_DOWN_PRED] = hor_down_##sz##_c; \
839  dsp->intra_pred[tx][VERT_LEFT_PRED] = vert_left_##sz##_c; \
840  dsp->intra_pred[tx][HOR_UP_PRED] = hor_up_##sz##_c; \
841  dsp->intra_pred[tx][TM_VP8_PRED] = tm_##sz##_c; \
842  dsp->intra_pred[tx][LEFT_DC_PRED] = dc_left_##sz##_c; \
843  dsp->intra_pred[tx][TOP_DC_PRED] = dc_top_##sz##_c; \
844  dsp->intra_pred[tx][DC_128_PRED] = dc_128_##sz##_c; \
845  dsp->intra_pred[tx][DC_127_PRED] = dc_127_##sz##_c; \
846  dsp->intra_pred[tx][DC_129_PRED] = dc_129_##sz##_c
847 
848  init_intra_pred(TX_4X4, 4x4);
849  init_intra_pred(TX_8X8, 8x8);
850  init_intra_pred(TX_16X16, 16x16);
851  init_intra_pred(TX_32X32, 32x32);
852 
853 #undef init_intra_pred
854 }
855 
856 #define itxfm_wrapper(type_a, type_b, sz, bits, has_dconly) \
857 static void type_a##_##type_b##_##sz##x##sz##_add_c(uint8_t *dst, \
858  ptrdiff_t stride, \
859  int16_t *block, int eob) \
860 { \
861  int i, j; \
862  int16_t tmp[sz * sz], out[sz]; \
863 \
864  if (has_dconly && eob == 1) { \
865  const int t = (((block[0] * 11585 + (1 << 13)) >> 14) \
866  * 11585 + (1 << 13)) >> 14; \
867  block[0] = 0; \
868  for (i = 0; i < sz; i++) { \
869  for (j = 0; j < sz; j++) \
870  dst[j * stride] = av_clip_uint8(dst[j * stride] + \
871  (bits ? \
872  (t + (1 << (bits - 1))) >> bits : \
873  t)); \
874  dst++; \
875  } \
876  return; \
877  } \
878 \
879  for (i = 0; i < sz; i++) \
880  type_a##sz##_1d(block + i, sz, tmp + i * sz, 0); \
881  memset(block, 0, sz * sz * sizeof(*block)); \
882  for (i = 0; i < sz; i++) { \
883  type_b##sz##_1d(tmp + i, sz, out, 1); \
884  for (j = 0; j < sz; j++) \
885  dst[j * stride] = av_clip_uint8(dst[j * stride] + \
886  (bits ? \
887  (out[j] + (1 << (bits - 1))) >> bits : \
888  out[j])); \
889  dst++; \
890  } \
891 }
892 
893 #define itxfm_wrap(sz, bits) \
894 itxfm_wrapper(idct, idct, sz, bits, 1) \
895 itxfm_wrapper(iadst, idct, sz, bits, 0) \
896 itxfm_wrapper(idct, iadst, sz, bits, 0) \
897 itxfm_wrapper(iadst, iadst, sz, bits, 0)
898 
899 #define IN(x) in[(x) * stride]
900 
901 static av_always_inline void idct4_1d(const int16_t *in, ptrdiff_t stride,
902  int16_t *out, int pass)
903 {
904  int t0, t1, t2, t3;
905 
906  t0 = ((IN(0) + IN(2)) * 11585 + (1 << 13)) >> 14;
907  t1 = ((IN(0) - IN(2)) * 11585 + (1 << 13)) >> 14;
908  t2 = (IN(1) * 6270 - IN(3) * 15137 + (1 << 13)) >> 14;
909  t3 = (IN(1) * 15137 + IN(3) * 6270 + (1 << 13)) >> 14;
910 
911  out[0] = t0 + t3;
912  out[1] = t1 + t2;
913  out[2] = t1 - t2;
914  out[3] = t0 - t3;
915 }
916 
917 static av_always_inline void iadst4_1d(const int16_t *in, ptrdiff_t stride,
918  int16_t *out, int pass)
919 {
920  int t0, t1, t2, t3;
921 
922  t0 = 5283 * IN(0) + 15212 * IN(2) + 9929 * IN(3);
923  t1 = 9929 * IN(0) - 5283 * IN(2) - 15212 * IN(3);
924  t2 = 13377 * (IN(0) - IN(2) + IN(3));
925  t3 = 13377 * IN(1);
926 
927  out[0] = (t0 + t3 + (1 << 13)) >> 14;
928  out[1] = (t1 + t3 + (1 << 13)) >> 14;
929  out[2] = (t2 + (1 << 13)) >> 14;
930  out[3] = (t0 + t1 - t3 + (1 << 13)) >> 14;
931 }
932 
934 
935 static av_always_inline void idct8_1d(const int16_t *in, ptrdiff_t stride,
936  int16_t *out, int pass)
937 {
938  int t0, t0a, t1, t1a, t2, t2a, t3, t3a, t4, t4a, t5, t5a, t6, t6a, t7, t7a;
939 
940  t0a = ((IN(0) + IN(4)) * 11585 + (1 << 13)) >> 14;
941  t1a = ((IN(0) - IN(4)) * 11585 + (1 << 13)) >> 14;
942  t2a = (IN(2) * 6270 - IN(6) * 15137 + (1 << 13)) >> 14;
943  t3a = (IN(2) * 15137 + IN(6) * 6270 + (1 << 13)) >> 14;
944  t4a = (IN(1) * 3196 - IN(7) * 16069 + (1 << 13)) >> 14;
945  t5a = (IN(5) * 13623 - IN(3) * 9102 + (1 << 13)) >> 14;
946  t6a = (IN(5) * 9102 + IN(3) * 13623 + (1 << 13)) >> 14;
947  t7a = (IN(1) * 16069 + IN(7) * 3196 + (1 << 13)) >> 14;
948 
949  t0 = t0a + t3a;
950  t1 = t1a + t2a;
951  t2 = t1a - t2a;
952  t3 = t0a - t3a;
953  t4 = t4a + t5a;
954  t5a = t4a - t5a;
955  t7 = t7a + t6a;
956  t6a = t7a - t6a;
957 
958  t5 = ((t6a - t5a) * 11585 + (1 << 13)) >> 14;
959  t6 = ((t6a + t5a) * 11585 + (1 << 13)) >> 14;
960 
961  out[0] = t0 + t7;
962  out[1] = t1 + t6;
963  out[2] = t2 + t5;
964  out[3] = t3 + t4;
965  out[4] = t3 - t4;
966  out[5] = t2 - t5;
967  out[6] = t1 - t6;
968  out[7] = t0 - t7;
969 }
970 
971 static av_always_inline void iadst8_1d(const int16_t *in, ptrdiff_t stride,
972  int16_t *out, int pass)
973 {
974  int t0, t0a, t1, t1a, t2, t2a, t3, t3a, t4, t4a, t5, t5a, t6, t6a, t7, t7a;
975 
976  t0a = 16305 * IN(7) + 1606 * IN(0);
977  t1a = 1606 * IN(7) - 16305 * IN(0);
978  t2a = 14449 * IN(5) + 7723 * IN(2);
979  t3a = 7723 * IN(5) - 14449 * IN(2);
980  t4a = 10394 * IN(3) + 12665 * IN(4);
981  t5a = 12665 * IN(3) - 10394 * IN(4);
982  t6a = 4756 * IN(1) + 15679 * IN(6);
983  t7a = 15679 * IN(1) - 4756 * IN(6);
984 
985  t0 = (t0a + t4a + (1 << 13)) >> 14;
986  t1 = (t1a + t5a + (1 << 13)) >> 14;
987  t2 = (t2a + t6a + (1 << 13)) >> 14;
988  t3 = (t3a + t7a + (1 << 13)) >> 14;
989  t4 = (t0a - t4a + (1 << 13)) >> 14;
990  t5 = (t1a - t5a + (1 << 13)) >> 14;
991  t6 = (t2a - t6a + (1 << 13)) >> 14;
992  t7 = (t3a - t7a + (1 << 13)) >> 14;
993 
994  t4a = 15137 * t4 + 6270 * t5;
995  t5a = 6270 * t4 - 15137 * t5;
996  t6a = 15137 * t7 - 6270 * t6;
997  t7a = 6270 * t7 + 15137 * t6;
998 
999  out[0] = t0 + t2;
1000  out[7] = -(t1 + t3);
1001  t2 = t0 - t2;
1002  t3 = t1 - t3;
1003 
1004  out[1] = -((t4a + t6a + (1 << 13)) >> 14);
1005  out[6] = (t5a + t7a + (1 << 13)) >> 14;
1006  t6 = (t4a - t6a + (1 << 13)) >> 14;
1007  t7 = (t5a - t7a + (1 << 13)) >> 14;
1008 
1009  out[3] = -(((t2 + t3) * 11585 + (1 << 13)) >> 14);
1010  out[4] = ((t2 - t3) * 11585 + (1 << 13)) >> 14;
1011  out[2] = ((t6 + t7) * 11585 + (1 << 13)) >> 14;
1012  out[5] = -(((t6 - t7) * 11585 + (1 << 13)) >> 14);
1013 }
1014 
1016 
1017 static av_always_inline void idct16_1d(const int16_t *in, ptrdiff_t stride,
1018  int16_t *out, int pass)
1019 {
1020  int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12, t13, t14, t15;
1021  int t0a, t1a, t2a, t3a, t4a, t5a, t6a, t7a;
1022  int t8a, t9a, t10a, t11a, t12a, t13a, t14a, t15a;
1023 
1024  t0a = ((IN(0) + IN(8)) * 11585 + (1 << 13)) >> 14;
1025  t1a = ((IN(0) - IN(8)) * 11585 + (1 << 13)) >> 14;
1026  t2a = (IN(4) * 6270 - IN(12) * 15137 + (1 << 13)) >> 14;
1027  t3a = (IN(4) * 15137 + IN(12) * 6270 + (1 << 13)) >> 14;
1028  t4a = (IN(2) * 3196 - IN(14) * 16069 + (1 << 13)) >> 14;
1029  t7a = (IN(2) * 16069 + IN(14) * 3196 + (1 << 13)) >> 14;
1030  t5a = (IN(10) * 13623 - IN(6) * 9102 + (1 << 13)) >> 14;
1031  t6a = (IN(10) * 9102 + IN(6) * 13623 + (1 << 13)) >> 14;
1032  t8a = (IN(1) * 1606 - IN(15) * 16305 + (1 << 13)) >> 14;
1033  t15a = (IN(1) * 16305 + IN(15) * 1606 + (1 << 13)) >> 14;
1034  t9a = (IN(9) * 12665 - IN(7) * 10394 + (1 << 13)) >> 14;
1035  t14a = (IN(9) * 10394 + IN(7) * 12665 + (1 << 13)) >> 14;
1036  t10a = (IN(5) * 7723 - IN(11) * 14449 + (1 << 13)) >> 14;
1037  t13a = (IN(5) * 14449 + IN(11) * 7723 + (1 << 13)) >> 14;
1038  t11a = (IN(13) * 15679 - IN(3) * 4756 + (1 << 13)) >> 14;
1039  t12a = (IN(13) * 4756 + IN(3) * 15679 + (1 << 13)) >> 14;
1040 
1041  t0 = t0a + t3a;
1042  t1 = t1a + t2a;
1043  t2 = t1a - t2a;
1044  t3 = t0a - t3a;
1045  t4 = t4a + t5a;
1046  t5 = t4a - t5a;
1047  t6 = t7a - t6a;
1048  t7 = t7a + t6a;
1049  t8 = t8a + t9a;
1050  t9 = t8a - t9a;
1051  t10 = t11a - t10a;
1052  t11 = t11a + t10a;
1053  t12 = t12a + t13a;
1054  t13 = t12a - t13a;
1055  t14 = t15a - t14a;
1056  t15 = t15a + t14a;
1057 
1058  t5a = ((t6 - t5) * 11585 + (1 << 13)) >> 14;
1059  t6a = ((t6 + t5) * 11585 + (1 << 13)) >> 14;
1060  t9a = ( t14 * 6270 - t9 * 15137 + (1 << 13)) >> 14;
1061  t14a = ( t14 * 15137 + t9 * 6270 + (1 << 13)) >> 14;
1062  t10a = (-(t13 * 15137 + t10 * 6270) + (1 << 13)) >> 14;
1063  t13a = ( t13 * 6270 - t10 * 15137 + (1 << 13)) >> 14;
1064 
1065  t0a = t0 + t7;
1066  t1a = t1 + t6a;
1067  t2a = t2 + t5a;
1068  t3a = t3 + t4;
1069  t4 = t3 - t4;
1070  t5 = t2 - t5a;
1071  t6 = t1 - t6a;
1072  t7 = t0 - t7;
1073  t8a = t8 + t11;
1074  t9 = t9a + t10a;
1075  t10 = t9a - t10a;
1076  t11a = t8 - t11;
1077  t12a = t15 - t12;
1078  t13 = t14a - t13a;
1079  t14 = t14a + t13a;
1080  t15a = t15 + t12;
1081 
1082  t10a = ((t13 - t10) * 11585 + (1 << 13)) >> 14;
1083  t13a = ((t13 + t10) * 11585 + (1 << 13)) >> 14;
1084  t11 = ((t12a - t11a) * 11585 + (1 << 13)) >> 14;
1085  t12 = ((t12a + t11a) * 11585 + (1 << 13)) >> 14;
1086 
1087  out[ 0] = t0a + t15a;
1088  out[ 1] = t1a + t14;
1089  out[ 2] = t2a + t13a;
1090  out[ 3] = t3a + t12;
1091  out[ 4] = t4 + t11;
1092  out[ 5] = t5 + t10a;
1093  out[ 6] = t6 + t9;
1094  out[ 7] = t7 + t8a;
1095  out[ 8] = t7 - t8a;
1096  out[ 9] = t6 - t9;
1097  out[10] = t5 - t10a;
1098  out[11] = t4 - t11;
1099  out[12] = t3a - t12;
1100  out[13] = t2a - t13a;
1101  out[14] = t1a - t14;
1102  out[15] = t0a - t15a;
1103 }
1104 
1105 static av_always_inline void iadst16_1d(const int16_t *in, ptrdiff_t stride,
1106  int16_t *out, int pass)
1107 {
1108  int t0, t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12, t13, t14, t15;
1109  int t0a, t1a, t2a, t3a, t4a, t5a, t6a, t7a;
1110  int t8a, t9a, t10a, t11a, t12a, t13a, t14a, t15a;
1111 
1112  t0 = IN(15) * 16364 + IN(0) * 804;
1113  t1 = IN(15) * 804 - IN(0) * 16364;
1114  t2 = IN(13) * 15893 + IN(2) * 3981;
1115  t3 = IN(13) * 3981 - IN(2) * 15893;
1116  t4 = IN(11) * 14811 + IN(4) * 7005;
1117  t5 = IN(11) * 7005 - IN(4) * 14811;
1118  t6 = IN(9) * 13160 + IN(6) * 9760;
1119  t7 = IN(9) * 9760 - IN(6) * 13160;
1120  t8 = IN(7) * 11003 + IN(8) * 12140;
1121  t9 = IN(7) * 12140 - IN(8) * 11003;
1122  t10 = IN(5) * 8423 + IN(10) * 14053;
1123  t11 = IN(5) * 14053 - IN(10) * 8423;
1124  t12 = IN(3) * 5520 + IN(12) * 15426;
1125  t13 = IN(3) * 15426 - IN(12) * 5520;
1126  t14 = IN(1) * 2404 + IN(14) * 16207;
1127  t15 = IN(1) * 16207 - IN(14) * 2404;
1128 
1129  t0a = (t0 + t8 + (1 << 13)) >> 14;
1130  t1a = (t1 + t9 + (1 << 13)) >> 14;
1131  t2a = (t2 + t10 + (1 << 13)) >> 14;
1132  t3a = (t3 + t11 + (1 << 13)) >> 14;
1133  t4a = (t4 + t12 + (1 << 13)) >> 14;
1134  t5a = (t5 + t13 + (1 << 13)) >> 14;
1135  t6a = (t6 + t14 + (1 << 13)) >> 14;
1136  t7a = (t7 + t15 + (1 << 13)) >> 14;
1137  t8a = (t0 - t8 + (1 << 13)) >> 14;
1138  t9a = (t1 - t9 + (1 << 13)) >> 14;
1139  t10a = (t2 - t10 + (1 << 13)) >> 14;
1140  t11a = (t3 - t11 + (1 << 13)) >> 14;
1141  t12a = (t4 - t12 + (1 << 13)) >> 14;
1142  t13a = (t5 - t13 + (1 << 13)) >> 14;
1143  t14a = (t6 - t14 + (1 << 13)) >> 14;
1144  t15a = (t7 - t15 + (1 << 13)) >> 14;
1145 
1146  t8 = t8a * 16069 + t9a * 3196;
1147  t9 = t8a * 3196 - t9a * 16069;
1148  t10 = t10a * 9102 + t11a * 13623;
1149  t11 = t10a * 13623 - t11a * 9102;
1150  t12 = t13a * 16069 - t12a * 3196;
1151  t13 = t13a * 3196 + t12a * 16069;
1152  t14 = t15a * 9102 - t14a * 13623;
1153  t15 = t15a * 13623 + t14a * 9102;
1154 
1155  t0 = t0a + t4a;
1156  t1 = t1a + t5a;
1157  t2 = t2a + t6a;
1158  t3 = t3a + t7a;
1159  t4 = t0a - t4a;
1160  t5 = t1a - t5a;
1161  t6 = t2a - t6a;
1162  t7 = t3a - t7a;
1163  t8a = (t8 + t12 + (1 << 13)) >> 14;
1164  t9a = (t9 + t13 + (1 << 13)) >> 14;
1165  t10a = (t10 + t14 + (1 << 13)) >> 14;
1166  t11a = (t11 + t15 + (1 << 13)) >> 14;
1167  t12a = (t8 - t12 + (1 << 13)) >> 14;
1168  t13a = (t9 - t13 + (1 << 13)) >> 14;
1169  t14a = (t10 - t14 + (1 << 13)) >> 14;
1170  t15a = (t11 - t15 + (1 << 13)) >> 14;
1171 
1172  t4a = t4 * 15137 + t5 * 6270;
1173  t5a = t4 * 6270 - t5 * 15137;
1174  t6a = t7 * 15137 - t6 * 6270;
1175  t7a = t7 * 6270 + t6 * 15137;
1176  t12 = t12a * 15137 + t13a * 6270;
1177  t13 = t12a * 6270 - t13a * 15137;
1178  t14 = t15a * 15137 - t14a * 6270;
1179  t15 = t15a * 6270 + t14a * 15137;
1180 
1181  out[ 0] = t0 + t2;
1182  out[15] = -(t1 + t3);
1183  t2a = t0 - t2;
1184  t3a = t1 - t3;
1185  out[ 3] = -((t4a + t6a + (1 << 13)) >> 14);
1186  out[12] = (t5a + t7a + (1 << 13)) >> 14;
1187  t6 = (t4a - t6a + (1 << 13)) >> 14;
1188  t7 = (t5a - t7a + (1 << 13)) >> 14;
1189  out[ 1] = -(t8a + t10a);
1190  out[14] = t9a + t11a;
1191  t10 = t8a - t10a;
1192  t11 = t9a - t11a;
1193  out[ 2] = (t12 + t14 + (1 << 13)) >> 14;
1194  out[13] = -((t13 + t15 + (1 << 13)) >> 14);
1195  t14a = (t12 - t14 + (1 << 13)) >> 14;
1196  t15a = (t13 - t15 + (1 << 13)) >> 14;
1197 
1198  out[ 7] = ((t2a + t3a) * -11585 + (1 << 13)) >> 14;
1199  out[ 8] = ((t2a - t3a) * 11585 + (1 << 13)) >> 14;
1200  out[ 4] = ((t7 + t6) * 11585 + (1 << 13)) >> 14;
1201  out[11] = ((t7 - t6) * 11585 + (1 << 13)) >> 14;
1202  out[ 6] = ((t11 + t10) * 11585 + (1 << 13)) >> 14;
1203  out[ 9] = ((t11 - t10) * 11585 + (1 << 13)) >> 14;
1204  out[ 5] = ((t14a + t15a) * -11585 + (1 << 13)) >> 14;
1205  out[10] = ((t14a - t15a) * 11585 + (1 << 13)) >> 14;
1206 }
1207 
1209 
1210 static av_always_inline void idct32_1d(const int16_t *in, ptrdiff_t stride,
1211  int16_t *out, int pass)
1212 {
1213  int t0a = ((IN(0) + IN(16)) * 11585 + (1 << 13)) >> 14;
1214  int t1a = ((IN(0) - IN(16)) * 11585 + (1 << 13)) >> 14;
1215  int t2a = (IN( 8) * 6270 - IN(24) * 15137 + (1 << 13)) >> 14;
1216  int t3a = (IN( 8) * 15137 + IN(24) * 6270 + (1 << 13)) >> 14;
1217  int t4a = (IN( 4) * 3196 - IN(28) * 16069 + (1 << 13)) >> 14;
1218  int t7a = (IN( 4) * 16069 + IN(28) * 3196 + (1 << 13)) >> 14;
1219  int t5a = (IN(20) * 13623 - IN(12) * 9102 + (1 << 13)) >> 14;
1220  int t6a = (IN(20) * 9102 + IN(12) * 13623 + (1 << 13)) >> 14;
1221  int t8a = (IN( 2) * 1606 - IN(30) * 16305 + (1 << 13)) >> 14;
1222  int t15a = (IN( 2) * 16305 + IN(30) * 1606 + (1 << 13)) >> 14;
1223  int t9a = (IN(18) * 12665 - IN(14) * 10394 + (1 << 13)) >> 14;
1224  int t14a = (IN(18) * 10394 + IN(14) * 12665 + (1 << 13)) >> 14;
1225  int t10a = (IN(10) * 7723 - IN(22) * 14449 + (1 << 13)) >> 14;
1226  int t13a = (IN(10) * 14449 + IN(22) * 7723 + (1 << 13)) >> 14;
1227  int t11a = (IN(26) * 15679 - IN( 6) * 4756 + (1 << 13)) >> 14;
1228  int t12a = (IN(26) * 4756 + IN( 6) * 15679 + (1 << 13)) >> 14;
1229  int t16a = (IN( 1) * 804 - IN(31) * 16364 + (1 << 13)) >> 14;
1230  int t31a = (IN( 1) * 16364 + IN(31) * 804 + (1 << 13)) >> 14;
1231  int t17a = (IN(17) * 12140 - IN(15) * 11003 + (1 << 13)) >> 14;
1232  int t30a = (IN(17) * 11003 + IN(15) * 12140 + (1 << 13)) >> 14;
1233  int t18a = (IN( 9) * 7005 - IN(23) * 14811 + (1 << 13)) >> 14;
1234  int t29a = (IN( 9) * 14811 + IN(23) * 7005 + (1 << 13)) >> 14;
1235  int t19a = (IN(25) * 15426 - IN( 7) * 5520 + (1 << 13)) >> 14;
1236  int t28a = (IN(25) * 5520 + IN( 7) * 15426 + (1 << 13)) >> 14;
1237  int t20a = (IN( 5) * 3981 - IN(27) * 15893 + (1 << 13)) >> 14;
1238  int t27a = (IN( 5) * 15893 + IN(27) * 3981 + (1 << 13)) >> 14;
1239  int t21a = (IN(21) * 14053 - IN(11) * 8423 + (1 << 13)) >> 14;
1240  int t26a = (IN(21) * 8423 + IN(11) * 14053 + (1 << 13)) >> 14;
1241  int t22a = (IN(13) * 9760 - IN(19) * 13160 + (1 << 13)) >> 14;
1242  int t25a = (IN(13) * 13160 + IN(19) * 9760 + (1 << 13)) >> 14;
1243  int t23a = (IN(29) * 16207 - IN( 3) * 2404 + (1 << 13)) >> 14;
1244  int t24a = (IN(29) * 2404 + IN( 3) * 16207 + (1 << 13)) >> 14;
1245 
1246  int t0 = t0a + t3a;
1247  int t1 = t1a + t2a;
1248  int t2 = t1a - t2a;
1249  int t3 = t0a - t3a;
1250  int t4 = t4a + t5a;
1251  int t5 = t4a - t5a;
1252  int t6 = t7a - t6a;
1253  int t7 = t7a + t6a;
1254  int t8 = t8a + t9a;
1255  int t9 = t8a - t9a;
1256  int t10 = t11a - t10a;
1257  int t11 = t11a + t10a;
1258  int t12 = t12a + t13a;
1259  int t13 = t12a - t13a;
1260  int t14 = t15a - t14a;
1261  int t15 = t15a + t14a;
1262  int t16 = t16a + t17a;
1263  int t17 = t16a - t17a;
1264  int t18 = t19a - t18a;
1265  int t19 = t19a + t18a;
1266  int t20 = t20a + t21a;
1267  int t21 = t20a - t21a;
1268  int t22 = t23a - t22a;
1269  int t23 = t23a + t22a;
1270  int t24 = t24a + t25a;
1271  int t25 = t24a - t25a;
1272  int t26 = t27a - t26a;
1273  int t27 = t27a + t26a;
1274  int t28 = t28a + t29a;
1275  int t29 = t28a - t29a;
1276  int t30 = t31a - t30a;
1277  int t31 = t31a + t30a;
1278 
1279  t5a = ((t6 - t5) * 11585 + (1 << 13)) >> 14;
1280  t6a = ((t6 + t5) * 11585 + (1 << 13)) >> 14;
1281  t9a = ( t14 * 6270 - t9 * 15137 + (1 << 13)) >> 14;
1282  t14a = ( t14 * 15137 + t9 * 6270 + (1 << 13)) >> 14;
1283  t10a = (-(t13 * 15137 + t10 * 6270) + (1 << 13)) >> 14;
1284  t13a = ( t13 * 6270 - t10 * 15137 + (1 << 13)) >> 14;
1285  t17a = ( t30 * 3196 - t17 * 16069 + (1 << 13)) >> 14;
1286  t30a = ( t30 * 16069 + t17 * 3196 + (1 << 13)) >> 14;
1287  t18a = (-(t29 * 16069 + t18 * 3196) + (1 << 13)) >> 14;
1288  t29a = ( t29 * 3196 - t18 * 16069 + (1 << 13)) >> 14;
1289  t21a = ( t26 * 13623 - t21 * 9102 + (1 << 13)) >> 14;
1290  t26a = ( t26 * 9102 + t21 * 13623 + (1 << 13)) >> 14;
1291  t22a = (-(t25 * 9102 + t22 * 13623) + (1 << 13)) >> 14;
1292  t25a = ( t25 * 13623 - t22 * 9102 + (1 << 13)) >> 14;
1293 
1294  t0a = t0 + t7;
1295  t1a = t1 + t6a;
1296  t2a = t2 + t5a;
1297  t3a = t3 + t4;
1298  t4a = t3 - t4;
1299  t5 = t2 - t5a;
1300  t6 = t1 - t6a;
1301  t7a = t0 - t7;
1302  t8a = t8 + t11;
1303  t9 = t9a + t10a;
1304  t10 = t9a - t10a;
1305  t11a = t8 - t11;
1306  t12a = t15 - t12;
1307  t13 = t14a - t13a;
1308  t14 = t14a + t13a;
1309  t15a = t15 + t12;
1310  t16a = t16 + t19;
1311  t17 = t17a + t18a;
1312  t18 = t17a - t18a;
1313  t19a = t16 - t19;
1314  t20a = t23 - t20;
1315  t21 = t22a - t21a;
1316  t22 = t22a + t21a;
1317  t23a = t23 + t20;
1318  t24a = t24 + t27;
1319  t25 = t25a + t26a;
1320  t26 = t25a - t26a;
1321  t27a = t24 - t27;
1322  t28a = t31 - t28;
1323  t29 = t30a - t29a;
1324  t30 = t30a + t29a;
1325  t31a = t31 + t28;
1326 
1327  t10a = ((t13 - t10) * 11585 + (1 << 13)) >> 14;
1328  t13a = ((t13 + t10) * 11585 + (1 << 13)) >> 14;
1329  t11 = ((t12a - t11a) * 11585 + (1 << 13)) >> 14;
1330  t12 = ((t12a + t11a) * 11585 + (1 << 13)) >> 14;
1331  t18a = ( t29 * 6270 - t18 * 15137 + (1 << 13)) >> 14;
1332  t29a = ( t29 * 15137 + t18 * 6270 + (1 << 13)) >> 14;
1333  t19 = ( t28a * 6270 - t19a * 15137 + (1 << 13)) >> 14;
1334  t28 = ( t28a * 15137 + t19a * 6270 + (1 << 13)) >> 14;
1335  t20 = (-(t27a * 15137 + t20a * 6270) + (1 << 13)) >> 14;
1336  t27 = ( t27a * 6270 - t20a * 15137 + (1 << 13)) >> 14;
1337  t21a = (-(t26 * 15137 + t21 * 6270) + (1 << 13)) >> 14;
1338  t26a = ( t26 * 6270 - t21 * 15137 + (1 << 13)) >> 14;
1339 
1340  t0 = t0a + t15a;
1341  t1 = t1a + t14;
1342  t2 = t2a + t13a;
1343  t3 = t3a + t12;
1344  t4 = t4a + t11;
1345  t5a = t5 + t10a;
1346  t6a = t6 + t9;
1347  t7 = t7a + t8a;
1348  t8 = t7a - t8a;
1349  t9a = t6 - t9;
1350  t10 = t5 - t10a;
1351  t11a = t4a - t11;
1352  t12a = t3a - t12;
1353  t13 = t2a - t13a;
1354  t14a = t1a - t14;
1355  t15 = t0a - t15a;
1356  t16 = t16a + t23a;
1357  t17a = t17 + t22;
1358  t18 = t18a + t21a;
1359  t19a = t19 + t20;
1360  t20a = t19 - t20;
1361  t21 = t18a - t21a;
1362  t22a = t17 - t22;
1363  t23 = t16a - t23a;
1364  t24 = t31a - t24a;
1365  t25a = t30 - t25;
1366  t26 = t29a - t26a;
1367  t27a = t28 - t27;
1368  t28a = t28 + t27;
1369  t29 = t29a + t26a;
1370  t30a = t30 + t25;
1371  t31 = t31a + t24a;
1372 
1373  t20 = ((t27a - t20a) * 11585 + (1 << 13)) >> 14;
1374  t27 = ((t27a + t20a) * 11585 + (1 << 13)) >> 14;
1375  t21a = ((t26 - t21 ) * 11585 + (1 << 13)) >> 14;
1376  t26a = ((t26 + t21 ) * 11585 + (1 << 13)) >> 14;
1377  t22 = ((t25a - t22a) * 11585 + (1 << 13)) >> 14;
1378  t25 = ((t25a + t22a) * 11585 + (1 << 13)) >> 14;
1379  t23a = ((t24 - t23 ) * 11585 + (1 << 13)) >> 14;
1380  t24a = ((t24 + t23 ) * 11585 + (1 << 13)) >> 14;
1381 
1382  out[ 0] = t0 + t31;
1383  out[ 1] = t1 + t30a;
1384  out[ 2] = t2 + t29;
1385  out[ 3] = t3 + t28a;
1386  out[ 4] = t4 + t27;
1387  out[ 5] = t5a + t26a;
1388  out[ 6] = t6a + t25;
1389  out[ 7] = t7 + t24a;
1390  out[ 8] = t8 + t23a;
1391  out[ 9] = t9a + t22;
1392  out[10] = t10 + t21a;
1393  out[11] = t11a + t20;
1394  out[12] = t12a + t19a;
1395  out[13] = t13 + t18;
1396  out[14] = t14a + t17a;
1397  out[15] = t15 + t16;
1398  out[16] = t15 - t16;
1399  out[17] = t14a - t17a;
1400  out[18] = t13 - t18;
1401  out[19] = t12a - t19a;
1402  out[20] = t11a - t20;
1403  out[21] = t10 - t21a;
1404  out[22] = t9a - t22;
1405  out[23] = t8 - t23a;
1406  out[24] = t7 - t24a;
1407  out[25] = t6a - t25;
1408  out[26] = t5a - t26a;
1409  out[27] = t4 - t27;
1410  out[28] = t3 - t28a;
1411  out[29] = t2 - t29;
1412  out[30] = t1 - t30a;
1413  out[31] = t0 - t31;
1414 }
1415 
1417 
1418 static av_always_inline void iwht4_1d(const int16_t *in, ptrdiff_t stride,
1419  int16_t *out, int pass)
1420 {
1421  int t0, t1, t2, t3, t4;
1422 
1423  if (pass == 0) {
1424  t0 = IN(0) >> 2;
1425  t1 = IN(3) >> 2;
1426  t2 = IN(1) >> 2;
1427  t3 = IN(2) >> 2;
1428  } else {
1429  t0 = IN(0);
1430  t1 = IN(3);
1431  t2 = IN(1);
1432  t3 = IN(2);
1433  }
1434 
1435  t0 += t2;
1436  t3 -= t1;
1437  t4 = (t0 - t3) >> 1;
1438  t1 = t4 - t1;
1439  t2 = t4 - t2;
1440  t0 -= t1;
1441  t3 += t2;
1442 
1443  out[0] = t0;
1444  out[1] = t1;
1445  out[2] = t2;
1446  out[3] = t3;
1447 }
1448 
1449 itxfm_wrapper(iwht, iwht, 4, 0, 0)
1450 
1451 #undef IN
1452 #undef itxfm_wrapper
1453 #undef itxfm_wrap
1454 
1455 static av_cold void vp9dsp_itxfm_init(VP9DSPContext *dsp)
1456 {
1457 #define init_itxfm(tx, sz) \
1458  dsp->itxfm_add[tx][DCT_DCT] = idct_idct_##sz##_add_c; \
1459  dsp->itxfm_add[tx][DCT_ADST] = iadst_idct_##sz##_add_c; \
1460  dsp->itxfm_add[tx][ADST_DCT] = idct_iadst_##sz##_add_c; \
1461  dsp->itxfm_add[tx][ADST_ADST] = iadst_iadst_##sz##_add_c
1462 
1463 #define init_idct(tx, nm) \
1464  dsp->itxfm_add[tx][DCT_DCT] = \
1465  dsp->itxfm_add[tx][ADST_DCT] = \
1466  dsp->itxfm_add[tx][DCT_ADST] = \
1467  dsp->itxfm_add[tx][ADST_ADST] = nm##_add_c
1468 
1469  init_itxfm(TX_4X4, 4x4);
1470  init_itxfm(TX_8X8, 8x8);
1471  init_itxfm(TX_16X16, 16x16);
1472  init_idct(TX_32X32, idct_idct_32x32);
1473  init_idct(4 /* lossless */, iwht_iwht_4x4);
1474 
1475 #undef init_itxfm
1476 #undef init_idct
1477 }
1478 
1479 static av_always_inline void loop_filter(uint8_t *dst, int E, int I, int H,
1480  ptrdiff_t stridea, ptrdiff_t strideb,
1481  int wd)
1482 {
1483  int i;
1484 
1485  for (i = 0; i < 8; i++, dst += stridea) {
1486  int p7, p6, p5, p4;
1487  int p3 = dst[strideb * -4], p2 = dst[strideb * -3];
1488  int p1 = dst[strideb * -2], p0 = dst[strideb * -1];
1489  int q0 = dst[strideb * +0], q1 = dst[strideb * +1];
1490  int q2 = dst[strideb * +2], q3 = dst[strideb * +3];
1491  int q4, q5, q6, q7;
1492  int fm = FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I &&
1493  FFABS(p1 - p0) <= I && FFABS(q1 - q0) <= I &&
1494  FFABS(q2 - q1) <= I && FFABS(q3 - q2) <= I &&
1495  FFABS(p0 - q0) * 2 + (FFABS(p1 - q1) >> 1) <= E;
1496  int flat8out, flat8in;
1497 
1498  if (!fm)
1499  continue;
1500 
1501  if (wd >= 16) {
1502  p7 = dst[strideb * -8];
1503  p6 = dst[strideb * -7];
1504  p5 = dst[strideb * -6];
1505  p4 = dst[strideb * -5];
1506  q4 = dst[strideb * +4];
1507  q5 = dst[strideb * +5];
1508  q6 = dst[strideb * +6];
1509  q7 = dst[strideb * +7];
1510 
1511  flat8out = FFABS(p7 - p0) <= 1 && FFABS(p6 - p0) <= 1 &&
1512  FFABS(p5 - p0) <= 1 && FFABS(p4 - p0) <= 1 &&
1513  FFABS(q4 - q0) <= 1 && FFABS(q5 - q0) <= 1 &&
1514  FFABS(q6 - q0) <= 1 && FFABS(q7 - q0) <= 1;
1515  }
1516 
1517  if (wd >= 8)
1518  flat8in = FFABS(p3 - p0) <= 1 && FFABS(p2 - p0) <= 1 &&
1519  FFABS(p1 - p0) <= 1 && FFABS(q1 - q0) <= 1 &&
1520  FFABS(q2 - q0) <= 1 && FFABS(q3 - q0) <= 1;
1521 
1522  if (wd >= 16 && flat8out && flat8in) {
1523  dst[strideb * -7] = (p7 + p7 + p7 + p7 + p7 + p7 + p7 + p6 * 2 +
1524  p5 + p4 + p3 + p2 + p1 + p0 + q0 + 8) >> 4;
1525  dst[strideb * -6] = (p7 + p7 + p7 + p7 + p7 + p7 + p6 + p5 * 2 +
1526  p4 + p3 + p2 + p1 + p0 + q0 + q1 + 8) >> 4;
1527  dst[strideb * -5] = (p7 + p7 + p7 + p7 + p7 + p6 + p5 + p4 * 2 +
1528  p3 + p2 + p1 + p0 + q0 + q1 + q2 + 8) >> 4;
1529  dst[strideb * -4] = (p7 + p7 + p7 + p7 + p6 + p5 + p4 + p3 * 2 +
1530  p2 + p1 + p0 + q0 + q1 + q2 + q3 + 8) >> 4;
1531  dst[strideb * -3] = (p7 + p7 + p7 + p6 + p5 + p4 + p3 + p2 * 2 +
1532  p1 + p0 + q0 + q1 + q2 + q3 + q4 + 8) >> 4;
1533  dst[strideb * -2] = (p7 + p7 + p6 + p5 + p4 + p3 + p2 + p1 * 2 +
1534  p0 + q0 + q1 + q2 + q3 + q4 + q5 + 8) >> 4;
1535  dst[strideb * -1] = (p7 + p6 + p5 + p4 + p3 + p2 + p1 + p0 * 2 +
1536  q0 + q1 + q2 + q3 + q4 + q5 + q6 + 8) >> 4;
1537  dst[strideb * +0] = (p6 + p5 + p4 + p3 + p2 + p1 + p0 + q0 * 2 +
1538  q1 + q2 + q3 + q4 + q5 + q6 + q7 + 8) >> 4;
1539  dst[strideb * +1] = (p5 + p4 + p3 + p2 + p1 + p0 + q0 + q1 * 2 +
1540  q2 + q3 + q4 + q5 + q6 + q7 + q7 + 8) >> 4;
1541  dst[strideb * +2] = (p4 + p3 + p2 + p1 + p0 + q0 + q1 + q2 * 2 +
1542  q3 + q4 + q5 + q6 + q7 + q7 + q7 + 8) >> 4;
1543  dst[strideb * +3] = (p3 + p2 + p1 + p0 + q0 + q1 + q2 + q3 * 2 +
1544  q4 + q5 + q6 + q7 + q7 + q7 + q7 + 8) >> 4;
1545  dst[strideb * +4] = (p2 + p1 + p0 + q0 + q1 + q2 + q3 + q4 * 2 +
1546  q5 + q6 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;
1547  dst[strideb * +5] = (p1 + p0 + q0 + q1 + q2 + q3 + q4 + q5 * 2 +
1548  q6 + q7 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;
1549  dst[strideb * +6] = (p0 + q0 + q1 + q2 + q3 + q4 + q5 + q6 * 2 +
1550  q7 + q7 + q7 + q7 + q7 + q7 + q7 + 8) >> 4;
1551  } else if (wd >= 8 && flat8in) {
1552  dst[strideb * -3] = (p3 + p3 + p3 + 2 * p2 + p1 + p0 + q0 + 4) >> 3;
1553  dst[strideb * -2] = (p3 + p3 + p2 + 2 * p1 + p0 + q0 + q1 + 4) >> 3;
1554  dst[strideb * -1] = (p3 + p2 + p1 + 2 * p0 + q0 + q1 + q2 + 4) >> 3;
1555  dst[strideb * +0] = (p2 + p1 + p0 + 2 * q0 + q1 + q2 + q3 + 4) >> 3;
1556  dst[strideb * +1] = (p1 + p0 + q0 + 2 * q1 + q2 + q3 + q3 + 4) >> 3;
1557  dst[strideb * +2] = (p0 + q0 + q1 + 2 * q2 + q3 + q3 + q3 + 4) >> 3;
1558  } else {
1559  int hev = FFABS(p1 - p0) > H || FFABS(q1 - q0) > H;
1560 
1561  if (hev) {
1562  int f = av_clip_int8(3 * (q0 - p0) + av_clip_int8(p1 - q1)), f1, f2;
1563 
1564  f1 = FFMIN(f + 4, 127) >> 3;
1565  f2 = FFMIN(f + 3, 127) >> 3;
1566 
1567  dst[strideb * -1] = av_clip_uint8(p0 + f2);
1568  dst[strideb * +0] = av_clip_uint8(q0 - f1);
1569  } else {
1570  int f = av_clip_int8(3 * (q0 - p0)), f1, f2;
1571 
1572  f1 = FFMIN(f + 4, 127) >> 3;
1573  f2 = FFMIN(f + 3, 127) >> 3;
1574 
1575  dst[strideb * -1] = av_clip_uint8(p0 + f2);
1576  dst[strideb * +0] = av_clip_uint8(q0 - f1);
1577 
1578  f = (f1 + 1) >> 1;
1579  dst[strideb * -2] = av_clip_uint8(p1 + f);
1580  dst[strideb * +1] = av_clip_uint8(q1 - f);
1581  }
1582  }
1583  }
1584 }
1585 
1586 #define lf_8_fn(dir, wd, stridea, strideb) \
1587 static void loop_filter_##dir##_##wd##_8_c(uint8_t *dst, \
1588  ptrdiff_t stride, \
1589  int E, int I, int H) \
1590 { \
1591  loop_filter(dst, E, I, H, stridea, strideb, wd); \
1592 }
1593 
1594 #define lf_8_fns(wd) \
1595 lf_8_fn(h, wd, stride, 1) \
1596 lf_8_fn(v, wd, 1, stride)
1597 
1599 lf_8_fns(8)
1600 lf_8_fns(16)
1601 
1602 #undef lf_8_fn
1603 #undef lf_8_fns
1604 
1605 #define lf_16_fn(dir, stridea) \
1606 static void loop_filter_##dir##_16_16_c(uint8_t *dst, \
1607  ptrdiff_t stride, \
1608  int E, int I, int H) \
1609 { \
1610  loop_filter_##dir##_16_8_c(dst, stride, E, I, H); \
1611  loop_filter_##dir##_16_8_c(dst + 8 * stridea, stride, E, I, H); \
1612 }
1613 
1614 lf_16_fn(h, stride)
1615 lf_16_fn(v, 1)
1616 
1617 #undef lf_16_fn
1618 
1619 #define lf_mix_fn(dir, wd1, wd2, stridea) \
1620 static void loop_filter_##dir##_##wd1##wd2##_16_c(uint8_t *dst, \
1621  ptrdiff_t stride, \
1622  int E, int I, int H) \
1623 { \
1624  loop_filter_##dir##_##wd1##_8_c(dst, stride, E & 0xff, I & 0xff, H & 0xff); \
1625  loop_filter_##dir##_##wd2##_8_c(dst + 8 * stridea, stride, E >> 8, I >> 8, H >> 8); \
1626 }
1627 
1628 #define lf_mix_fns(wd1, wd2) \
1629 lf_mix_fn(h, wd1, wd2, stride) \
1630 lf_mix_fn(v, wd1, wd2, 1)
1631 
1632 lf_mix_fns(4, 4)
1633 lf_mix_fns(4, 8)
1634 lf_mix_fns(8, 4)
1635 lf_mix_fns(8, 8)
1636 
1637 #undef lf_mix_fn
1638 #undef lf_mix_fns
1639 
1640 static av_cold void vp9dsp_loopfilter_init(VP9DSPContext *dsp)
1641 {
1642  dsp->loop_filter_8[0][0] = loop_filter_h_4_8_c;
1643  dsp->loop_filter_8[0][1] = loop_filter_v_4_8_c;
1644  dsp->loop_filter_8[1][0] = loop_filter_h_8_8_c;
1645  dsp->loop_filter_8[1][1] = loop_filter_v_8_8_c;
1646  dsp->loop_filter_8[2][0] = loop_filter_h_16_8_c;
1647  dsp->loop_filter_8[2][1] = loop_filter_v_16_8_c;
1648 
1649  dsp->loop_filter_16[0] = loop_filter_h_16_16_c;
1650  dsp->loop_filter_16[1] = loop_filter_v_16_16_c;
1651 
1652  dsp->loop_filter_mix2[0][0][0] = loop_filter_h_44_16_c;
1653  dsp->loop_filter_mix2[0][0][1] = loop_filter_v_44_16_c;
1654  dsp->loop_filter_mix2[0][1][0] = loop_filter_h_48_16_c;
1655  dsp->loop_filter_mix2[0][1][1] = loop_filter_v_48_16_c;
1656  dsp->loop_filter_mix2[1][0][0] = loop_filter_h_84_16_c;
1657  dsp->loop_filter_mix2[1][0][1] = loop_filter_v_84_16_c;
1658  dsp->loop_filter_mix2[1][1][0] = loop_filter_h_88_16_c;
1659  dsp->loop_filter_mix2[1][1][1] = loop_filter_v_88_16_c;
1660 }
1661 
1662 static av_always_inline void copy_c(uint8_t *dst, ptrdiff_t dst_stride,
1663  const uint8_t *src, ptrdiff_t src_stride,
1664  int w, int h)
1665 {
1666  do {
1667  memcpy(dst, src, w);
1668 
1669  dst += dst_stride;
1670  src += src_stride;
1671  } while (--h);
1672 }
1673 
1674 static av_always_inline void avg_c(uint8_t *dst, ptrdiff_t dst_stride,
1675  const uint8_t *src, ptrdiff_t src_stride,
1676  int w, int h)
1677 {
1678  do {
1679  int x;
1680 
1681  for (x = 0; x < w; x += 4)
1682  AV_WN32A(&dst[x], rnd_avg32(AV_RN32A(&dst[x]), AV_RN32(&src[x])));
1683 
1684  dst += dst_stride;
1685  src += src_stride;
1686  } while (--h);
1687 }
1688 
1689 #define fpel_fn(type, sz) \
1690 static void type##sz##_c(uint8_t *dst, ptrdiff_t dst_stride, \
1691  const uint8_t *src, ptrdiff_t src_stride, \
1692  int h, int mx, int my) \
1693 { \
1694  type##_c(dst, dst_stride, src, src_stride, sz, h); \
1695 }
1696 
1697 #define copy_avg_fn(sz) \
1698 fpel_fn(copy, sz) \
1699 fpel_fn(avg, sz)
1700 
1702 copy_avg_fn(32)
1703 copy_avg_fn(16)
1704 copy_avg_fn(8)
1705 copy_avg_fn(4)
1706 
1707 #undef fpel_fn
1708 #undef copy_avg_fn
1709 
1710 static const int8_t vp9_subpel_filters[3][15][8] = {
1711  [FILTER_8TAP_REGULAR] = {
1712  { 0, 1, -5, 126, 8, -3, 1, 0 },
1713  { -1, 3, -10, 122, 18, -6, 2, 0 },
1714  { -1, 4, -13, 118, 27, -9, 3, -1 },
1715  { -1, 4, -16, 112, 37, -11, 4, -1 },
1716  { -1, 5, -18, 105, 48, -14, 4, -1 },
1717  { -1, 5, -19, 97, 58, -16, 5, -1 },
1718  { -1, 6, -19, 88, 68, -18, 5, -1 },
1719  { -1, 6, -19, 78, 78, -19, 6, -1 },
1720  { -1, 5, -18, 68, 88, -19, 6, -1 },
1721  { -1, 5, -16, 58, 97, -19, 5, -1 },
1722  { -1, 4, -14, 48, 105, -18, 5, -1 },
1723  { -1, 4, -11, 37, 112, -16, 4, -1 },
1724  { -1, 3, -9, 27, 118, -13, 4, -1 },
1725  { 0, 2, -6, 18, 122, -10, 3, -1 },
1726  { 0, 1, -3, 8, 126, -5, 1, 0 },
1727  }, [FILTER_8TAP_SHARP] = {
1728  { -1, 3, -7, 127, 8, -3, 1, 0 },
1729  { -2, 5, -13, 125, 17, -6, 3, -1 },
1730  { -3, 7, -17, 121, 27, -10, 5, -2 },
1731  { -4, 9, -20, 115, 37, -13, 6, -2 },
1732  { -4, 10, -23, 108, 48, -16, 8, -3 },
1733  { -4, 10, -24, 100, 59, -19, 9, -3 },
1734  { -4, 11, -24, 90, 70, -21, 10, -4 },
1735  { -4, 11, -23, 80, 80, -23, 11, -4 },
1736  { -4, 10, -21, 70, 90, -24, 11, -4 },
1737  { -3, 9, -19, 59, 100, -24, 10, -4 },
1738  { -3, 8, -16, 48, 108, -23, 10, -4 },
1739  { -2, 6, -13, 37, 115, -20, 9, -4 },
1740  { -2, 5, -10, 27, 121, -17, 7, -3 },
1741  { -1, 3, -6, 17, 125, -13, 5, -2 },
1742  { 0, 1, -3, 8, 127, -7, 3, -1 },
1743  }, [FILTER_8TAP_SMOOTH] = {
1744  { -3, -1, 32, 64, 38, 1, -3, 0 },
1745  { -2, -2, 29, 63, 41, 2, -3, 0 },
1746  { -2, -2, 26, 63, 43, 4, -4, 0 },
1747  { -2, -3, 24, 62, 46, 5, -4, 0 },
1748  { -2, -3, 21, 60, 49, 7, -4, 0 },
1749  { -1, -4, 18, 59, 51, 9, -4, 0 },
1750  { -1, -4, 16, 57, 53, 12, -4, -1 },
1751  { -1, -4, 14, 55, 55, 14, -4, -1 },
1752  { -1, -4, 12, 53, 57, 16, -4, -1 },
1753  { 0, -4, 9, 51, 59, 18, -4, -1 },
1754  { 0, -4, 7, 49, 60, 21, -3, -2 },
1755  { 0, -4, 5, 46, 62, 24, -3, -2 },
1756  { 0, -4, 4, 43, 63, 26, -2, -2 },
1757  { 0, -3, 2, 41, 63, 29, -2, -2 },
1758  { 0, -3, 1, 38, 64, 32, -1, -3 },
1759  }
1760 };
1761 
1762 #define FILTER_8TAP(src, x, F, stride) \
1763  av_clip_uint8((F[0] * src[x + -3 * stride] + \
1764  F[1] * src[x + -2 * stride] + \
1765  F[2] * src[x + -1 * stride] + \
1766  F[3] * src[x + +0 * stride] + \
1767  F[4] * src[x + +1 * stride] + \
1768  F[5] * src[x + +2 * stride] + \
1769  F[6] * src[x + +3 * stride] + \
1770  F[7] * src[x + +4 * stride] + 64) >> 7)
1771 
1772 static av_always_inline void do_8tap_1d_c(uint8_t *dst, ptrdiff_t dst_stride,
1773  const uint8_t *src, ptrdiff_t src_stride,
1774  int w, int h, ptrdiff_t ds,
1775  const int8_t *filter, int avg)
1776 {
1777  do {
1778  int x;
1779 
1780  for (x = 0; x < w; x++)
1781  if (avg) {
1782  dst[x] = (dst[x] + FILTER_8TAP(src, x, filter, ds) + 1) >> 1;
1783  } else {
1784  dst[x] = FILTER_8TAP(src, x, filter, ds);
1785  }
1786 
1787  dst += dst_stride;
1788  src += src_stride;
1789  } while (--h);
1790 }
1791 
1792 #define filter_8tap_1d_fn(opn, opa, dir, ds) \
1793 static av_noinline void opn##_8tap_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \
1794  const uint8_t *src, ptrdiff_t src_stride, \
1795  int w, int h, const int8_t *filter) \
1796 { \
1797  do_8tap_1d_c(dst, dst_stride, src, src_stride, w, h, ds, filter, opa); \
1798 }
1799 
1800 filter_8tap_1d_fn(put, 0, v, src_stride)
1801 filter_8tap_1d_fn(put, 0, h, 1)
1802 filter_8tap_1d_fn(avg, 1, v, src_stride)
1803 filter_8tap_1d_fn(avg, 1, h, 1)
1804 
1805 #undef filter_8tap_1d_fn
1806 
1807 static av_always_inline void do_8tap_2d_c(uint8_t *dst, ptrdiff_t dst_stride,
1808  const uint8_t *src, ptrdiff_t src_stride,
1809  int w, int h, const int8_t *filterx,
1810  const int8_t *filtery, int avg)
1811 {
1812  int tmp_h = h + 7;
1813  uint8_t tmp[64 * 71], *tmp_ptr = tmp;
1814 
1815  src -= src_stride * 3;
1816  do {
1817  int x;
1818 
1819  for (x = 0; x < w; x++)
1820  tmp_ptr[x] = FILTER_8TAP(src, x, filterx, 1);
1821 
1822  tmp_ptr += 64;
1823  src += src_stride;
1824  } while (--tmp_h);
1825 
1826  tmp_ptr = tmp + 64 * 3;
1827  do {
1828  int x;
1829 
1830  for (x = 0; x < w; x++)
1831  if (avg) {
1832  dst[x] = (dst[x] + FILTER_8TAP(tmp_ptr, x, filtery, 64) + 1) >> 1;
1833  } else {
1834  dst[x] = FILTER_8TAP(tmp_ptr, x, filtery, 64);
1835  }
1836 
1837  tmp_ptr += 64;
1838  dst += dst_stride;
1839  } while (--h);
1840 }
1841 
1842 #define filter_8tap_2d_fn(opn, opa) \
1843 static av_noinline void opn##_8tap_2d_hv_c(uint8_t *dst, ptrdiff_t dst_stride, \
1844  const uint8_t *src, ptrdiff_t src_stride, \
1845  int w, int h, const int8_t *filterx, \
1846  const int8_t *filtery) \
1847 { \
1848  do_8tap_2d_c(dst, dst_stride, src, src_stride, w, h, filterx, filtery, opa); \
1849 }
1850 
1852 filter_8tap_2d_fn(avg, 1)
1853 
1854 #undef filter_8tap_2d_fn
1855 
1856 #undef FILTER_8TAP
1857 
1858 #define filter_fn_1d(sz, dir, dir_m, type, type_idx, avg) \
1859 static void avg##_8tap_##type##_##sz##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \
1860  const uint8_t *src, ptrdiff_t src_stride, \
1861  int h, int mx, int my) \
1862 { \
1863  avg##_8tap_1d_##dir##_c(dst, dst_stride, src, src_stride, sz, h, \
1864  vp9_subpel_filters[type_idx][dir_m - 1]); \
1865 }
1866 
1867 #define filter_fn_2d(sz, type, type_idx, avg) \
1868 static void avg##_8tap_##type##_##sz##hv_c(uint8_t *dst, ptrdiff_t dst_stride, \
1869  const uint8_t *src, ptrdiff_t src_stride, \
1870  int h, int mx, int my) \
1871 { \
1872  avg##_8tap_2d_hv_c(dst, dst_stride, src, src_stride, sz, h, \
1873  vp9_subpel_filters[type_idx][mx - 1], \
1874  vp9_subpel_filters[type_idx][my - 1]); \
1875 }
1876 
1877 #define FILTER_BILIN(src, x, mxy, stride) \
1878  (src[x] + ((mxy * (src[x + stride] - src[x]) + 8) >> 4))
1879 
1880 static av_always_inline void do_bilin_1d_c(uint8_t *dst, ptrdiff_t dst_stride,
1881  const uint8_t *src, ptrdiff_t src_stride,
1882  int w, int h, ptrdiff_t ds, int mxy, int avg)
1883 {
1884  do {
1885  int x;
1886 
1887  for (x = 0; x < w; x++)
1888  if (avg) {
1889  dst[x] = (dst[x] + FILTER_BILIN(src, x, mxy, ds) + 1) >> 1;
1890  } else {
1891  dst[x] = FILTER_BILIN(src, x, mxy, ds);
1892  }
1893 
1894  dst += dst_stride;
1895  src += src_stride;
1896  } while (--h);
1897 }
1898 
1899 #define bilin_1d_fn(opn, opa, dir, ds) \
1900 static av_noinline void opn##_bilin_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \
1901  const uint8_t *src, ptrdiff_t src_stride, \
1902  int w, int h, int mxy) \
1903 { \
1904  do_bilin_1d_c(dst, dst_stride, src, src_stride, w, h, ds, mxy, opa); \
1905 }
1906 
1907 bilin_1d_fn(put, 0, v, src_stride)
1908 bilin_1d_fn(put, 0, h, 1)
1909 bilin_1d_fn(avg, 1, v, src_stride)
1910 bilin_1d_fn(avg, 1, h, 1)
1911 
1912 #undef bilin_1d_fn
1913 
1914 static av_always_inline void do_bilin_2d_c(uint8_t *dst, ptrdiff_t dst_stride,
1915  const uint8_t *src, ptrdiff_t src_stride,
1916  int w, int h, int mx, int my, int avg)
1917 {
1918  uint8_t tmp[64 * 65], *tmp_ptr = tmp;
1919  int tmp_h = h + 1;
1920 
1921  do {
1922  int x;
1923 
1924  for (x = 0; x < w; x++)
1925  tmp_ptr[x] = FILTER_BILIN(src, x, mx, 1);
1926 
1927  tmp_ptr += 64;
1928  src += src_stride;
1929  } while (--tmp_h);
1930 
1931  tmp_ptr = tmp;
1932  do {
1933  int x;
1934 
1935  for (x = 0; x < w; x++)
1936  if (avg) {
1937  dst[x] = (dst[x] + FILTER_BILIN(tmp_ptr, x, my, 64) + 1) >> 1;
1938  } else {
1939  dst[x] = FILTER_BILIN(tmp_ptr, x, my, 64);
1940  }
1941 
1942  tmp_ptr += 64;
1943  dst += dst_stride;
1944  } while (--h);
1945 }
1946 
1947 #define bilin_2d_fn(opn, opa) \
1948 static av_noinline void opn##_bilin_2d_hv_c(uint8_t *dst, ptrdiff_t dst_stride, \
1949  const uint8_t *src, ptrdiff_t src_stride, \
1950  int w, int h, int mx, int my) \
1951 { \
1952  do_bilin_2d_c(dst, dst_stride, src, src_stride, w, h, mx, my, opa); \
1953 }
1954 
1956 bilin_2d_fn(avg, 1)
1957 
1958 #undef bilin_2d_fn
1959 
1960 #undef FILTER_BILIN
1961 
1962 #define bilinf_fn_1d(sz, dir, dir_m, avg) \
1963 static void avg##_bilin_##sz##dir##_c(uint8_t *dst, ptrdiff_t dst_stride, \
1964  const uint8_t *src, ptrdiff_t src_stride, \
1965  int h, int mx, int my) \
1966 { \
1967  avg##_bilin_1d_##dir##_c(dst, dst_stride, src, src_stride, sz, h, dir_m); \
1968 }
1969 
1970 #define bilinf_fn_2d(sz, avg) \
1971 static void avg##_bilin_##sz##hv_c(uint8_t *dst, ptrdiff_t dst_stride, \
1972  const uint8_t *src, ptrdiff_t src_stride, \
1973  int h, int mx, int my) \
1974 { \
1975  avg##_bilin_2d_hv_c(dst, dst_stride, src, src_stride, sz, h, mx, my); \
1976 }
1977 
1978 #define filter_fn(sz, avg) \
1979 filter_fn_1d(sz, h, mx, regular, FILTER_8TAP_REGULAR, avg) \
1980 filter_fn_1d(sz, v, my, regular, FILTER_8TAP_REGULAR, avg) \
1981 filter_fn_2d(sz, regular, FILTER_8TAP_REGULAR, avg) \
1982 filter_fn_1d(sz, h, mx, smooth, FILTER_8TAP_SMOOTH, avg) \
1983 filter_fn_1d(sz, v, my, smooth, FILTER_8TAP_SMOOTH, avg) \
1984 filter_fn_2d(sz, smooth, FILTER_8TAP_SMOOTH, avg) \
1985 filter_fn_1d(sz, h, mx, sharp, FILTER_8TAP_SHARP, avg) \
1986 filter_fn_1d(sz, v, my, sharp, FILTER_8TAP_SHARP, avg) \
1987 filter_fn_2d(sz, sharp, FILTER_8TAP_SHARP, avg) \
1988 bilinf_fn_1d(sz, h, mx, avg) \
1989 bilinf_fn_1d(sz, v, my, avg) \
1990 bilinf_fn_2d(sz, avg)
1991 
1992 #define filter_fn_set(avg) \
1993 filter_fn(64, avg) \
1994 filter_fn(32, avg) \
1995 filter_fn(16, avg) \
1996 filter_fn(8, avg) \
1997 filter_fn(4, avg)
1998 
1999 filter_fn_set(put)
2000 filter_fn_set(avg)
2001 
2002 #undef filter_fn
2003 #undef filter_fn_set
2004 #undef filter_fn_1d
2005 #undef filter_fn_2d
2006 #undef bilinf_fn_1d
2007 #undef bilinf_fn_2d
2008 
2009 static av_cold void vp9dsp_mc_init(VP9DSPContext *dsp)
2010 {
2011 #define init_fpel(idx1, idx2, sz, type) \
2012  dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = type##sz##_c; \
2013  dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = type##sz##_c; \
2014  dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = type##sz##_c; \
2015  dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = type##sz##_c
2016 
2017 #define init_copy_avg(idx, sz) \
2018  init_fpel(idx, 0, sz, copy); \
2019  init_fpel(idx, 1, sz, avg)
2020 
2021  init_copy_avg(0, 64);
2022  init_copy_avg(1, 32);
2023  init_copy_avg(2, 16);
2024  init_copy_avg(3, 8);
2025  init_copy_avg(4, 4);
2026 
2027 #undef init_copy_avg
2028 #undef init_fpel
2029 
2030 #define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type) \
2031  dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_c; \
2032  dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_c; \
2033  dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_c; \
2034  dsp->mc[idx1][FILTER_BILINEAR ][idx2][idxh][idxv] = type##_bilin_##sz##dir##_c
2035 
2036 #define init_subpel2(idx, idxh, idxv, dir, type) \
2037  init_subpel1(0, idx, idxh, idxv, 64, dir, type); \
2038  init_subpel1(1, idx, idxh, idxv, 32, dir, type); \
2039  init_subpel1(2, idx, idxh, idxv, 16, dir, type); \
2040  init_subpel1(3, idx, idxh, idxv, 8, dir, type); \
2041  init_subpel1(4, idx, idxh, idxv, 4, dir, type)
2042 
2043 #define init_subpel3(idx, type) \
2044  init_subpel2(idx, 1, 1, hv, type); \
2045  init_subpel2(idx, 0, 1, v, type); \
2046  init_subpel2(idx, 1, 0, h, type)
2047 
2048  init_subpel3(0, put);
2049  init_subpel3(1, avg);
2050 
2051 #undef init_subpel1
2052 #undef init_subpel2
2053 #undef init_subpel3
2054 }
2055 
2057 {
2058  vp9dsp_intrapred_init(dsp);
2059  vp9dsp_itxfm_init(dsp);
2060  vp9dsp_loopfilter_init(dsp);
2061  vp9dsp_mc_init(dsp);
2062 
2063  if (ARCH_X86) ff_vp9dsp_init_x86(dsp);
2064 }