avcodec/pthread_frame: fix setting hwaccel with threads and get_format()
[ffmpeg.git] / libavcodec / vp8dsp.c
1 /*
2  * Copyright (C) 2010 David Conrad
3  * Copyright (C) 2010 Ronald S. Bultje
4  * Copyright (C) 2014 Peter Ross
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 /**
24  * @file
25  * VP8 compatible video decoder
26  */
27
28 #include "libavutil/common.h"
29 #include "libavutil/intreadwrite.h"
30
31 #include "mathops.h"
32 #include "vp8dsp.h"
33
34 #define MK_IDCT_DC_ADD4_C(name)                                               \
35 static void name ## _idct_dc_add4uv_c(uint8_t *dst, int16_t block[4][16],     \
36                                       ptrdiff_t stride)                       \
37 {                                                                             \
38     name ## _idct_dc_add_c(dst + stride * 0 + 0, block[0], stride);           \
39     name ## _idct_dc_add_c(dst + stride * 0 + 4, block[1], stride);           \
40     name ## _idct_dc_add_c(dst + stride * 4 + 0, block[2], stride);           \
41     name ## _idct_dc_add_c(dst + stride * 4 + 4, block[3], stride);           \
42 }                                                                             \
43                                                                               \
44 static void name ## _idct_dc_add4y_c(uint8_t *dst, int16_t block[4][16],      \
45                                      ptrdiff_t stride)                        \
46 {                                                                             \
47     name ## _idct_dc_add_c(dst +  0, block[0], stride);                       \
48     name ## _idct_dc_add_c(dst +  4, block[1], stride);                       \
49     name ## _idct_dc_add_c(dst +  8, block[2], stride);                       \
50     name ## _idct_dc_add_c(dst + 12, block[3], stride);                       \
51 }
52
53 #if CONFIG_VP7_DECODER
54 static void vp7_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
55 {
56     int i, a1, b1, c1, d1;
57     int16_t tmp[16];
58
59     for (i = 0; i < 4; i++) {
60         a1 = (dc[i * 4 + 0] + dc[i * 4 + 2]) * 23170;
61         b1 = (dc[i * 4 + 0] - dc[i * 4 + 2]) * 23170;
62         c1 = dc[i * 4 + 1] * 12540 - dc[i * 4 + 3] * 30274;
63         d1 = dc[i * 4 + 1] * 30274 + dc[i * 4 + 3] * 12540;
64         tmp[i * 4 + 0] = (a1 + d1) >> 14;
65         tmp[i * 4 + 3] = (a1 - d1) >> 14;
66         tmp[i * 4 + 1] = (b1 + c1) >> 14;
67         tmp[i * 4 + 2] = (b1 - c1) >> 14;
68     }
69
70     for (i = 0; i < 4; i++) {
71         a1 = (tmp[i + 0] + tmp[i + 8]) * 23170;
72         b1 = (tmp[i + 0] - tmp[i + 8]) * 23170;
73         c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274;
74         d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540;
75         AV_ZERO64(dc + i * 4);
76         block[0][i][0] = (a1 + d1 + 0x20000) >> 18;
77         block[3][i][0] = (a1 - d1 + 0x20000) >> 18;
78         block[1][i][0] = (b1 + c1 + 0x20000) >> 18;
79         block[2][i][0] = (b1 - c1 + 0x20000) >> 18;
80     }
81 }
82
83 static void vp7_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
84 {
85     int i, val = (23170 * (23170 * dc[0] >> 14) + 0x20000) >> 18;
86     dc[0] = 0;
87
88     for (i = 0; i < 4; i++) {
89         block[i][0][0] = val;
90         block[i][1][0] = val;
91         block[i][2][0] = val;
92         block[i][3][0] = val;
93     }
94 }
95
96 static void vp7_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
97 {
98     int i, a1, b1, c1, d1;
99     int16_t tmp[16];
100
101     for (i = 0; i < 4; i++) {
102         a1 = (block[i * 4 + 0] + block[i * 4 + 2]) * 23170;
103         b1 = (block[i * 4 + 0] - block[i * 4 + 2]) * 23170;
104         c1 = block[i * 4 + 1] * 12540 - block[i * 4 + 3] * 30274;
105         d1 = block[i * 4 + 1] * 30274 + block[i * 4 + 3] * 12540;
106         AV_ZERO64(block + i * 4);
107         tmp[i * 4 + 0] = (a1 + d1) >> 14;
108         tmp[i * 4 + 3] = (a1 - d1) >> 14;
109         tmp[i * 4 + 1] = (b1 + c1) >> 14;
110         tmp[i * 4 + 2] = (b1 - c1) >> 14;
111     }
112
113     for (i = 0; i < 4; i++) {
114         a1 = (tmp[i + 0] + tmp[i + 8]) * 23170;
115         b1 = (tmp[i + 0] - tmp[i + 8]) * 23170;
116         c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274;
117         d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540;
118         dst[0 * stride + i] = av_clip_uint8(dst[0 * stride + i] +
119                                             ((a1 + d1 + 0x20000) >> 18));
120         dst[3 * stride + i] = av_clip_uint8(dst[3 * stride + i] +
121                                             ((a1 - d1 + 0x20000) >> 18));
122         dst[1 * stride + i] = av_clip_uint8(dst[1 * stride + i] +
123                                             ((b1 + c1 + 0x20000) >> 18));
124         dst[2 * stride + i] = av_clip_uint8(dst[2 * stride + i] +
125                                             ((b1 - c1 + 0x20000) >> 18));
126     }
127 }
128
129 static void vp7_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
130 {
131     int i, dc = (23170 * (23170 * block[0] >> 14) + 0x20000) >> 18;
132     block[0] = 0;
133
134     for (i = 0; i < 4; i++) {
135         dst[0] = av_clip_uint8(dst[0] + dc);
136         dst[1] = av_clip_uint8(dst[1] + dc);
137         dst[2] = av_clip_uint8(dst[2] + dc);
138         dst[3] = av_clip_uint8(dst[3] + dc);
139         dst   += stride;
140     }
141 }
142
143 MK_IDCT_DC_ADD4_C(vp7)
144 #endif /* CONFIG_VP7_DECODER */
145
146 // TODO: Maybe add dequant
147 #if CONFIG_VP8_DECODER
148 static void vp8_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
149 {
150     int i, t0, t1, t2, t3;
151
152     for (i = 0; i < 4; i++) {
153         t0 = dc[0 * 4 + i] + dc[3 * 4 + i];
154         t1 = dc[1 * 4 + i] + dc[2 * 4 + i];
155         t2 = dc[1 * 4 + i] - dc[2 * 4 + i];
156         t3 = dc[0 * 4 + i] - dc[3 * 4 + i];
157
158         dc[0 * 4 + i] = t0 + t1;
159         dc[1 * 4 + i] = t3 + t2;
160         dc[2 * 4 + i] = t0 - t1;
161         dc[3 * 4 + i] = t3 - t2;
162     }
163
164     for (i = 0; i < 4; i++) {
165         t0 = dc[i * 4 + 0] + dc[i * 4 + 3] + 3; // rounding
166         t1 = dc[i * 4 + 1] + dc[i * 4 + 2];
167         t2 = dc[i * 4 + 1] - dc[i * 4 + 2];
168         t3 = dc[i * 4 + 0] - dc[i * 4 + 3] + 3; // rounding
169         AV_ZERO64(dc + i * 4);
170
171         block[i][0][0] = (t0 + t1) >> 3;
172         block[i][1][0] = (t3 + t2) >> 3;
173         block[i][2][0] = (t0 - t1) >> 3;
174         block[i][3][0] = (t3 - t2) >> 3;
175     }
176 }
177
178 static void vp8_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
179 {
180     int i, val = (dc[0] + 3) >> 3;
181     dc[0] = 0;
182
183     for (i = 0; i < 4; i++) {
184         block[i][0][0] = val;
185         block[i][1][0] = val;
186         block[i][2][0] = val;
187         block[i][3][0] = val;
188     }
189 }
190
191 #define MUL_20091(a) ((((a) * 20091) >> 16) + (a))
192 #define MUL_35468(a)  (((a) * 35468) >> 16)
193
194 static void vp8_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
195 {
196     int i, t0, t1, t2, t3;
197     int16_t tmp[16];
198
199     for (i = 0; i < 4; i++) {
200         t0 = block[0 * 4 + i] + block[2 * 4 + i];
201         t1 = block[0 * 4 + i] - block[2 * 4 + i];
202         t2 = MUL_35468(block[1 * 4 + i]) - MUL_20091(block[3 * 4 + i]);
203         t3 = MUL_20091(block[1 * 4 + i]) + MUL_35468(block[3 * 4 + i]);
204         block[0 * 4 + i] = 0;
205         block[1 * 4 + i] = 0;
206         block[2 * 4 + i] = 0;
207         block[3 * 4 + i] = 0;
208
209         tmp[i * 4 + 0] = t0 + t3;
210         tmp[i * 4 + 1] = t1 + t2;
211         tmp[i * 4 + 2] = t1 - t2;
212         tmp[i * 4 + 3] = t0 - t3;
213     }
214
215     for (i = 0; i < 4; i++) {
216         t0 = tmp[0 * 4 + i] + tmp[2 * 4 + i];
217         t1 = tmp[0 * 4 + i] - tmp[2 * 4 + i];
218         t2 = MUL_35468(tmp[1 * 4 + i]) - MUL_20091(tmp[3 * 4 + i]);
219         t3 = MUL_20091(tmp[1 * 4 + i]) + MUL_35468(tmp[3 * 4 + i]);
220
221         dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
222         dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
223         dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
224         dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
225         dst   += stride;
226     }
227 }
228
229 static void vp8_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
230 {
231     int i, dc = (block[0] + 4) >> 3;
232     block[0] = 0;
233
234     for (i = 0; i < 4; i++) {
235         dst[0] = av_clip_uint8(dst[0] + dc);
236         dst[1] = av_clip_uint8(dst[1] + dc);
237         dst[2] = av_clip_uint8(dst[2] + dc);
238         dst[3] = av_clip_uint8(dst[3] + dc);
239         dst   += stride;
240     }
241 }
242
243 MK_IDCT_DC_ADD4_C(vp8)
244 #endif /* CONFIG_VP8_DECODER */
245
246 // because I like only having two parameters to pass functions...
247 #define LOAD_PIXELS                                                           \
248     int av_unused p3 = p[-4 * stride];                                        \
249     int av_unused p2 = p[-3 * stride];                                        \
250     int av_unused p1 = p[-2 * stride];                                        \
251     int av_unused p0 = p[-1 * stride];                                        \
252     int av_unused q0 = p[ 0 * stride];                                        \
253     int av_unused q1 = p[ 1 * stride];                                        \
254     int av_unused q2 = p[ 2 * stride];                                        \
255     int av_unused q3 = p[ 3 * stride];
256
257 #define clip_int8(n) (cm[(n) + 0x80] - 0x80)
258
259 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride,
260                                            int is4tap, int is_vp7)
261 {
262     LOAD_PIXELS
263     int a, f1, f2;
264     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
265
266     a = 3 * (q0 - p0);
267
268     if (is4tap)
269         a += clip_int8(p1 - q1);
270
271     a = clip_int8(a);
272
273     // We deviate from the spec here with c(a+3) >> 3
274     // since that's what libvpx does.
275     f1 = FFMIN(a + 4, 127) >> 3;
276
277     if (is_vp7)
278         f2 = f1 - ((a & 7) == 4);
279     else
280         f2 = FFMIN(a + 3, 127) >> 3;
281
282     // Despite what the spec says, we do need to clamp here to
283     // be bitexact with libvpx.
284     p[-1 * stride] = cm[p0 + f2];
285     p[ 0 * stride] = cm[q0 - f1];
286
287     // only used for _inner on blocks without high edge variance
288     if (!is4tap) {
289         a              = (f1 + 1) >> 1;
290         p[-2 * stride] = cm[p1 + a];
291         p[ 1 * stride] = cm[q1 - a];
292     }
293 }
294
295 static av_always_inline void vp7_filter_common(uint8_t *p, ptrdiff_t stride,
296                                                int is4tap)
297 {
298     filter_common(p, stride, is4tap, IS_VP7);
299 }
300
301 static av_always_inline void vp8_filter_common(uint8_t *p, ptrdiff_t stride,
302                                                int is4tap)
303 {
304     filter_common(p, stride, is4tap, IS_VP8);
305 }
306
307 static av_always_inline int vp7_simple_limit(uint8_t *p, ptrdiff_t stride,
308                                              int flim)
309 {
310     LOAD_PIXELS
311     return FFABS(p0 - q0) <= flim;
312 }
313
314 static av_always_inline int vp8_simple_limit(uint8_t *p, ptrdiff_t stride,
315                                              int flim)
316 {
317     LOAD_PIXELS
318     return 2 * FFABS(p0 - q0) + (FFABS(p1 - q1) >> 1) <= flim;
319 }
320
321 /**
322  * E - limit at the macroblock edge
323  * I - limit for interior difference
324  */
325 #define NORMAL_LIMIT(vpn)                                                     \
326 static av_always_inline int vp ## vpn ## _normal_limit(uint8_t *p,            \
327                                                        ptrdiff_t stride,      \
328                                                        int E, int I)          \
329 {                                                                             \
330     LOAD_PIXELS                                                               \
331     return vp ## vpn ## _simple_limit(p, stride, E) &&                        \
332            FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I &&                      \
333            FFABS(p1 - p0) <= I && FFABS(q3 - q2) <= I &&                      \
334            FFABS(q2 - q1) <= I && FFABS(q1 - q0) <= I;                        \
335 }
336
337 NORMAL_LIMIT(7)
338 NORMAL_LIMIT(8)
339
340 // high edge variance
341 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh)
342 {
343     LOAD_PIXELS
344     return FFABS(p1 - p0) > thresh || FFABS(q1 - q0) > thresh;
345 }
346
347 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride)
348 {
349     int a0, a1, a2, w;
350     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
351
352     LOAD_PIXELS
353
354     w = clip_int8(p1 - q1);
355     w = clip_int8(w + 3 * (q0 - p0));
356
357     a0 = (27 * w + 63) >> 7;
358     a1 = (18 * w + 63) >> 7;
359     a2 =  (9 * w + 63) >> 7;
360
361     p[-3 * stride] = cm[p2 + a2];
362     p[-2 * stride] = cm[p1 + a1];
363     p[-1 * stride] = cm[p0 + a0];
364     p[ 0 * stride] = cm[q0 - a0];
365     p[ 1 * stride] = cm[q1 - a1];
366     p[ 2 * stride] = cm[q2 - a2];
367 }
368
369 #define LOOP_FILTER(vpn, dir, size, stridea, strideb, maybe_inline)           \
370 static maybe_inline                                                           \
371 void vpn ## _ ## dir ## _loop_filter ## size ## _c(uint8_t *dst,              \
372                                                    ptrdiff_t stride,          \
373                                                    int flim_E, int flim_I,    \
374                                                    int hev_thresh)            \
375 {                                                                             \
376     int i;                                                                    \
377     for (i = 0; i < size; i++)                                                \
378         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
379                                  flim_E, flim_I)) {                           \
380             if (hev(dst + i * stridea, strideb, hev_thresh))                  \
381                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
382             else                                                              \
383                 filter_mbedge(dst + i * stridea, strideb);                    \
384         }                                                                     \
385 }                                                                             \
386                                                                               \
387 static maybe_inline                                                           \
388 void vpn ## _ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst,        \
389                                                          ptrdiff_t stride,    \
390                                                          int flim_E,          \
391                                                          int flim_I,          \
392                                                          int hev_thresh)      \
393 {                                                                             \
394     int i;                                                                    \
395     for (i = 0; i < size; i++)                                                \
396         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
397                                  flim_E, flim_I)) {                           \
398             int hv = hev(dst + i * stridea, strideb, hev_thresh);             \
399             if (hv)                                                           \
400                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
401             else                                                              \
402                 vpn ## _filter_common(dst + i * stridea, strideb, 0);         \
403         }                                                                     \
404 }
405
406 #define UV_LOOP_FILTER(vpn, dir, stridea, strideb)                            \
407 LOOP_FILTER(vpn, dir, 8, stridea, strideb, av_always_inline)                  \
408 static void vpn ## _ ## dir ## _loop_filter8uv_c(uint8_t *dstU,               \
409                                                  uint8_t *dstV,               \
410                                                  ptrdiff_t stride, int fE,    \
411                                                  int fI, int hev_thresh)      \
412 {                                                                             \
413     vpn ## _ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);     \
414     vpn ## _ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);     \
415 }                                                                             \
416                                                                               \
417 static void vpn ## _ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU,         \
418                                                        uint8_t *dstV,         \
419                                                        ptrdiff_t stride,      \
420                                                        int fE, int fI,        \
421                                                        int hev_thresh)        \
422 {                                                                             \
423     vpn ## _ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI,            \
424                                              hev_thresh);                     \
425     vpn ## _ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI,            \
426                                              hev_thresh);                     \
427 }
428
429 #define LOOP_FILTER_SIMPLE(vpn)                                               \
430 static void vpn ## _v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
431                                            int flim)                          \
432 {                                                                             \
433     int i;                                                                    \
434     for (i = 0; i < 16; i++)                                                  \
435         if (vpn ## _simple_limit(dst + i, stride, flim))                      \
436             vpn ## _filter_common(dst + i, stride, 1);                        \
437 }                                                                             \
438                                                                               \
439 static void vpn ## _h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
440                                            int flim)                          \
441 {                                                                             \
442     int i;                                                                    \
443     for (i = 0; i < 16; i++)                                                  \
444         if (vpn ## _simple_limit(dst + i * stride, 1, flim))                  \
445             vpn ## _filter_common(dst + i * stride, 1, 1);                    \
446 }
447
448 #define LOOP_FILTERS(vpn)                \
449     LOOP_FILTER(vpn, v, 16, 1, stride, ) \
450     LOOP_FILTER(vpn, h, 16, stride, 1, ) \
451     UV_LOOP_FILTER(vpn, v, 1, stride)    \
452     UV_LOOP_FILTER(vpn, h, stride, 1)    \
453     LOOP_FILTER_SIMPLE(vpn)              \
454
455 static const uint8_t subpel_filters[7][6] = {
456     { 0,  6, 123,  12,  1, 0 },
457     { 2, 11, 108,  36,  8, 1 },
458     { 0,  9,  93,  50,  6, 0 },
459     { 3, 16,  77,  77, 16, 3 },
460     { 0,  6,  50,  93,  9, 0 },
461     { 1,  8,  36, 108, 11, 2 },
462     { 0,  1,  12, 123,  6, 0 },
463 };
464
465 #define PUT_PIXELS(WIDTH)                                                     \
466 static void put_vp8_pixels ## WIDTH ## _c(uint8_t *dst, ptrdiff_t dststride,  \
467                                           uint8_t *src, ptrdiff_t srcstride,  \
468                                           int h, int x, int y)                \
469 {                                                                             \
470     int i;                                                                    \
471     for (i = 0; i < h; i++, dst += dststride, src += srcstride)               \
472         memcpy(dst, src, WIDTH);                                              \
473 }
474
475 PUT_PIXELS(16)
476 PUT_PIXELS(8)
477 PUT_PIXELS(4)
478
479 #define FILTER_6TAP(src, F, stride)                                           \
480     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
481         F[0] * src[x - 2 * stride] + F[3] * src[x + 1 * stride] -             \
482         F[4] * src[x + 2 * stride] + F[5] * src[x + 3 * stride] + 64) >> 7]
483
484 #define FILTER_4TAP(src, F, stride)                                           \
485     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
486         F[3] * src[x + 1 * stride] - F[4] * src[x + 2 * stride] + 64) >> 7]
487
488 #define VP8_EPEL_H(SIZE, TAPS)                                                \
489 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst,            \
490                                                      ptrdiff_t dststride,     \
491                                                      uint8_t *src,            \
492                                                      ptrdiff_t srcstride,     \
493                                                      int h, int mx, int my)   \
494 {                                                                             \
495     const uint8_t *filter = subpel_filters[mx - 1];                           \
496     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
497     int x, y;                                                                 \
498     for (y = 0; y < h; y++) {                                                 \
499         for (x = 0; x < SIZE; x++)                                            \
500             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1);                  \
501         dst += dststride;                                                     \
502         src += srcstride;                                                     \
503     }                                                                         \
504 }
505
506 #define VP8_EPEL_V(SIZE, TAPS)                                                \
507 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst,            \
508                                                      ptrdiff_t dststride,     \
509                                                      uint8_t *src,            \
510                                                      ptrdiff_t srcstride,     \
511                                                      int h, int mx, int my)   \
512 {                                                                             \
513     const uint8_t *filter = subpel_filters[my - 1];                           \
514     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
515     int x, y;                                                                 \
516     for (y = 0; y < h; y++) {                                                 \
517         for (x = 0; x < SIZE; x++)                                            \
518             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride);          \
519         dst += dststride;                                                     \
520         src += srcstride;                                                     \
521     }                                                                         \
522 }
523
524 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS)                                       \
525 static void                                                                   \
526 put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst,         \
527                                                         ptrdiff_t dststride,  \
528                                                         uint8_t *src,         \
529                                                         ptrdiff_t srcstride,  \
530                                                         int h, int mx,        \
531                                                         int my)               \
532 {                                                                             \
533     const uint8_t *filter = subpel_filters[mx - 1];                           \
534     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
535     int x, y;                                                                 \
536     uint8_t tmp_array[(2 * SIZE + VTAPS - 1) * SIZE];                         \
537     uint8_t *tmp = tmp_array;                                                 \
538     src -= (2 - (VTAPS == 4)) * srcstride;                                    \
539                                                                               \
540     for (y = 0; y < h + VTAPS - 1; y++) {                                     \
541         for (x = 0; x < SIZE; x++)                                            \
542             tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1);                 \
543         tmp += SIZE;                                                          \
544         src += srcstride;                                                     \
545     }                                                                         \
546     tmp    = tmp_array + (2 - (VTAPS == 4)) * SIZE;                           \
547     filter = subpel_filters[my - 1];                                          \
548                                                                               \
549     for (y = 0; y < h; y++) {                                                 \
550         for (x = 0; x < SIZE; x++)                                            \
551             dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE);              \
552         dst += dststride;                                                     \
553         tmp += SIZE;                                                          \
554     }                                                                         \
555 }
556
557 VP8_EPEL_H(16, 4)
558 VP8_EPEL_H(8,  4)
559 VP8_EPEL_H(4,  4)
560 VP8_EPEL_H(16, 6)
561 VP8_EPEL_H(8,  6)
562 VP8_EPEL_H(4,  6)
563 VP8_EPEL_V(16, 4)
564 VP8_EPEL_V(8,  4)
565 VP8_EPEL_V(4,  4)
566 VP8_EPEL_V(16, 6)
567 VP8_EPEL_V(8,  6)
568 VP8_EPEL_V(4,  6)
569
570 VP8_EPEL_HV(16, 4, 4)
571 VP8_EPEL_HV(8,  4, 4)
572 VP8_EPEL_HV(4,  4, 4)
573 VP8_EPEL_HV(16, 4, 6)
574 VP8_EPEL_HV(8,  4, 6)
575 VP8_EPEL_HV(4,  4, 6)
576 VP8_EPEL_HV(16, 6, 4)
577 VP8_EPEL_HV(8,  6, 4)
578 VP8_EPEL_HV(4,  6, 4)
579 VP8_EPEL_HV(16, 6, 6)
580 VP8_EPEL_HV(8,  6, 6)
581 VP8_EPEL_HV(4,  6, 6)
582
583 #define VP8_BILINEAR(SIZE)                                                    \
584 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t dstride, \
585                                              uint8_t *src, ptrdiff_t sstride, \
586                                              int h, int mx, int my)           \
587 {                                                                             \
588     int a = 8 - mx, b = mx;                                                   \
589     int x, y;                                                                 \
590     for (y = 0; y < h; y++) {                                                 \
591         for (x = 0; x < SIZE; x++)                                            \
592             dst[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
593         dst += dstride;                                                       \
594         src += sstride;                                                       \
595     }                                                                         \
596 }                                                                             \
597                                                                               \
598 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t dstride, \
599                                              uint8_t *src, ptrdiff_t sstride, \
600                                              int h, int mx, int my)           \
601 {                                                                             \
602     int c = 8 - my, d = my;                                                   \
603     int x, y;                                                                 \
604     for (y = 0; y < h; y++) {                                                 \
605         for (x = 0; x < SIZE; x++)                                            \
606             dst[x] = (c * src[x] + d * src[x + sstride] + 4) >> 3;            \
607         dst += dstride;                                                       \
608         src += sstride;                                                       \
609     }                                                                         \
610 }                                                                             \
611                                                                               \
612 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst,                   \
613                                               ptrdiff_t dstride,              \
614                                               uint8_t *src,                   \
615                                               ptrdiff_t sstride,              \
616                                               int h, int mx, int my)          \
617 {                                                                             \
618     int a = 8 - mx, b = mx;                                                   \
619     int c = 8 - my, d = my;                                                   \
620     int x, y;                                                                 \
621     uint8_t tmp_array[(2 * SIZE + 1) * SIZE];                                 \
622     uint8_t *tmp = tmp_array;                                                 \
623     for (y = 0; y < h + 1; y++) {                                             \
624         for (x = 0; x < SIZE; x++)                                            \
625             tmp[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
626         tmp += SIZE;                                                          \
627         src += sstride;                                                       \
628     }                                                                         \
629     tmp = tmp_array;                                                          \
630     for (y = 0; y < h; y++) {                                                 \
631         for (x = 0; x < SIZE; x++)                                            \
632             dst[x] = (c * tmp[x] + d * tmp[x + SIZE] + 4) >> 3;               \
633         dst += dstride;                                                       \
634         tmp += SIZE;                                                          \
635     }                                                                         \
636 }
637
638 VP8_BILINEAR(16)
639 VP8_BILINEAR(8)
640 VP8_BILINEAR(4)
641
642 #define VP78_MC_FUNC(IDX, SIZE)                                               \
643     dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c;   \
644     dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c;  \
645     dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c;  \
646     dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c;  \
647     dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
648     dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
649     dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c;  \
650     dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
651     dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
652
653 #define VP78_BILINEAR_MC_FUNC(IDX, SIZE)                                      \
654     dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels   ## SIZE ## _c; \
655     dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
656     dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
657     dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
658     dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
659     dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
660     dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
661     dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
662     dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
663
664 av_cold void ff_vp78dsp_init(VP8DSPContext *dsp)
665 {
666     VP78_MC_FUNC(0, 16);
667     VP78_MC_FUNC(1, 8);
668     VP78_MC_FUNC(2, 4);
669
670     VP78_BILINEAR_MC_FUNC(0, 16);
671     VP78_BILINEAR_MC_FUNC(1, 8);
672     VP78_BILINEAR_MC_FUNC(2, 4);
673
674     if (ARCH_ARM)
675         ff_vp78dsp_init_arm(dsp);
676     if (ARCH_PPC)
677         ff_vp78dsp_init_ppc(dsp);
678     if (ARCH_X86)
679         ff_vp78dsp_init_x86(dsp);
680 }
681
682 #if CONFIG_VP7_DECODER
683 LOOP_FILTERS(vp7)
684
685 av_cold void ff_vp7dsp_init(VP8DSPContext *dsp)
686 {
687     dsp->vp8_luma_dc_wht    = vp7_luma_dc_wht_c;
688     dsp->vp8_luma_dc_wht_dc = vp7_luma_dc_wht_dc_c;
689     dsp->vp8_idct_add       = vp7_idct_add_c;
690     dsp->vp8_idct_dc_add    = vp7_idct_dc_add_c;
691     dsp->vp8_idct_dc_add4y  = vp7_idct_dc_add4y_c;
692     dsp->vp8_idct_dc_add4uv = vp7_idct_dc_add4uv_c;
693
694     dsp->vp8_v_loop_filter16y = vp7_v_loop_filter16_c;
695     dsp->vp8_h_loop_filter16y = vp7_h_loop_filter16_c;
696     dsp->vp8_v_loop_filter8uv = vp7_v_loop_filter8uv_c;
697     dsp->vp8_h_loop_filter8uv = vp7_h_loop_filter8uv_c;
698
699     dsp->vp8_v_loop_filter16y_inner = vp7_v_loop_filter16_inner_c;
700     dsp->vp8_h_loop_filter16y_inner = vp7_h_loop_filter16_inner_c;
701     dsp->vp8_v_loop_filter8uv_inner = vp7_v_loop_filter8uv_inner_c;
702     dsp->vp8_h_loop_filter8uv_inner = vp7_h_loop_filter8uv_inner_c;
703
704     dsp->vp8_v_loop_filter_simple = vp7_v_loop_filter_simple_c;
705     dsp->vp8_h_loop_filter_simple = vp7_h_loop_filter_simple_c;
706 }
707 #endif /* CONFIG_VP7_DECODER */
708
709 #if CONFIG_VP8_DECODER
710 LOOP_FILTERS(vp8)
711
712 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
713 {
714     dsp->vp8_luma_dc_wht    = vp8_luma_dc_wht_c;
715     dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
716     dsp->vp8_idct_add       = vp8_idct_add_c;
717     dsp->vp8_idct_dc_add    = vp8_idct_dc_add_c;
718     dsp->vp8_idct_dc_add4y  = vp8_idct_dc_add4y_c;
719     dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
720
721     dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
722     dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
723     dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
724     dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
725
726     dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
727     dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
728     dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
729     dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
730
731     dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
732     dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
733
734     if (ARCH_ARM)
735         ff_vp8dsp_init_arm(dsp);
736     if (ARCH_X86)
737         ff_vp8dsp_init_x86(dsp);
738 }
739 #endif /* CONFIG_VP8_DECODER */