avcodec/vp8dsp: Fixes: runtime error: signed integer overflow: 1330143360 - -10230405...
[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;
99     unsigned a1, b1, c1, d1;
100     int16_t tmp[16];
101
102     for (i = 0; i < 4; i++) {
103         a1 = (block[i * 4 + 0] + block[i * 4 + 2]) * 23170;
104         b1 = (block[i * 4 + 0] - block[i * 4 + 2]) * 23170;
105         c1 = block[i * 4 + 1] * 12540 - block[i * 4 + 3] * 30274;
106         d1 = block[i * 4 + 1] * 30274 + block[i * 4 + 3] * 12540;
107         AV_ZERO64(block + i * 4);
108         tmp[i * 4 + 0] = (int)(a1 + d1) >> 14;
109         tmp[i * 4 + 3] = (int)(a1 - d1) >> 14;
110         tmp[i * 4 + 1] = (int)(b1 + c1) >> 14;
111         tmp[i * 4 + 2] = (int)(b1 - c1) >> 14;
112     }
113
114     for (i = 0; i < 4; i++) {
115         a1 = (tmp[i + 0] + tmp[i + 8]) * 23170;
116         b1 = (tmp[i + 0] - tmp[i + 8]) * 23170;
117         c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274;
118         d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540;
119         dst[0 * stride + i] = av_clip_uint8(dst[0 * stride + i] +
120                                             ((int)(a1 + d1 + 0x20000) >> 18));
121         dst[3 * stride + i] = av_clip_uint8(dst[3 * stride + i] +
122                                             ((int)(a1 - d1 + 0x20000) >> 18));
123         dst[1 * stride + i] = av_clip_uint8(dst[1 * stride + i] +
124                                             ((int)(b1 + c1 + 0x20000) >> 18));
125         dst[2 * stride + i] = av_clip_uint8(dst[2 * stride + i] +
126                                             ((int)(b1 - c1 + 0x20000) >> 18));
127     }
128 }
129
130 static void vp7_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
131 {
132     int i, dc = (23170 * (23170 * block[0] >> 14) + 0x20000) >> 18;
133     block[0] = 0;
134
135     for (i = 0; i < 4; i++) {
136         dst[0] = av_clip_uint8(dst[0] + dc);
137         dst[1] = av_clip_uint8(dst[1] + dc);
138         dst[2] = av_clip_uint8(dst[2] + dc);
139         dst[3] = av_clip_uint8(dst[3] + dc);
140         dst   += stride;
141     }
142 }
143
144 MK_IDCT_DC_ADD4_C(vp7)
145 #endif /* CONFIG_VP7_DECODER */
146
147 // TODO: Maybe add dequant
148 #if CONFIG_VP8_DECODER
149 static void vp8_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
150 {
151     int i, t0, t1, t2, t3;
152
153     for (i = 0; i < 4; i++) {
154         t0 = dc[0 * 4 + i] + dc[3 * 4 + i];
155         t1 = dc[1 * 4 + i] + dc[2 * 4 + i];
156         t2 = dc[1 * 4 + i] - dc[2 * 4 + i];
157         t3 = dc[0 * 4 + i] - dc[3 * 4 + i];
158
159         dc[0 * 4 + i] = t0 + t1;
160         dc[1 * 4 + i] = t3 + t2;
161         dc[2 * 4 + i] = t0 - t1;
162         dc[3 * 4 + i] = t3 - t2;
163     }
164
165     for (i = 0; i < 4; i++) {
166         t0 = dc[i * 4 + 0] + dc[i * 4 + 3] + 3; // rounding
167         t1 = dc[i * 4 + 1] + dc[i * 4 + 2];
168         t2 = dc[i * 4 + 1] - dc[i * 4 + 2];
169         t3 = dc[i * 4 + 0] - dc[i * 4 + 3] + 3; // rounding
170         AV_ZERO64(dc + i * 4);
171
172         block[i][0][0] = (t0 + t1) >> 3;
173         block[i][1][0] = (t3 + t2) >> 3;
174         block[i][2][0] = (t0 - t1) >> 3;
175         block[i][3][0] = (t3 - t2) >> 3;
176     }
177 }
178
179 static void vp8_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
180 {
181     int i, val = (dc[0] + 3) >> 3;
182     dc[0] = 0;
183
184     for (i = 0; i < 4; i++) {
185         block[i][0][0] = val;
186         block[i][1][0] = val;
187         block[i][2][0] = val;
188         block[i][3][0] = val;
189     }
190 }
191
192 #define MUL_20091(a) ((((a) * 20091) >> 16) + (a))
193 #define MUL_35468(a)  (((a) * 35468) >> 16)
194
195 static void vp8_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
196 {
197     int i, t0, t1, t2, t3;
198     int16_t tmp[16];
199
200     for (i = 0; i < 4; i++) {
201         t0 = block[0 * 4 + i] + block[2 * 4 + i];
202         t1 = block[0 * 4 + i] - block[2 * 4 + i];
203         t2 = MUL_35468(block[1 * 4 + i]) - MUL_20091(block[3 * 4 + i]);
204         t3 = MUL_20091(block[1 * 4 + i]) + MUL_35468(block[3 * 4 + i]);
205         block[0 * 4 + i] = 0;
206         block[1 * 4 + i] = 0;
207         block[2 * 4 + i] = 0;
208         block[3 * 4 + i] = 0;
209
210         tmp[i * 4 + 0] = t0 + t3;
211         tmp[i * 4 + 1] = t1 + t2;
212         tmp[i * 4 + 2] = t1 - t2;
213         tmp[i * 4 + 3] = t0 - t3;
214     }
215
216     for (i = 0; i < 4; i++) {
217         t0 = tmp[0 * 4 + i] + tmp[2 * 4 + i];
218         t1 = tmp[0 * 4 + i] - tmp[2 * 4 + i];
219         t2 = MUL_35468(tmp[1 * 4 + i]) - MUL_20091(tmp[3 * 4 + i]);
220         t3 = MUL_20091(tmp[1 * 4 + i]) + MUL_35468(tmp[3 * 4 + i]);
221
222         dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
223         dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
224         dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
225         dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
226         dst   += stride;
227     }
228 }
229
230 static void vp8_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
231 {
232     int i, dc = (block[0] + 4) >> 3;
233     block[0] = 0;
234
235     for (i = 0; i < 4; i++) {
236         dst[0] = av_clip_uint8(dst[0] + dc);
237         dst[1] = av_clip_uint8(dst[1] + dc);
238         dst[2] = av_clip_uint8(dst[2] + dc);
239         dst[3] = av_clip_uint8(dst[3] + dc);
240         dst   += stride;
241     }
242 }
243
244 MK_IDCT_DC_ADD4_C(vp8)
245 #endif /* CONFIG_VP8_DECODER */
246
247 // because I like only having two parameters to pass functions...
248 #define LOAD_PIXELS                                                           \
249     int av_unused p3 = p[-4 * stride];                                        \
250     int av_unused p2 = p[-3 * stride];                                        \
251     int av_unused p1 = p[-2 * stride];                                        \
252     int av_unused p0 = p[-1 * stride];                                        \
253     int av_unused q0 = p[ 0 * stride];                                        \
254     int av_unused q1 = p[ 1 * stride];                                        \
255     int av_unused q2 = p[ 2 * stride];                                        \
256     int av_unused q3 = p[ 3 * stride];
257
258 #define clip_int8(n) (cm[(n) + 0x80] - 0x80)
259
260 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride,
261                                            int is4tap, int is_vp7)
262 {
263     LOAD_PIXELS
264     int a, f1, f2;
265     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
266
267     a = 3 * (q0 - p0);
268
269     if (is4tap)
270         a += clip_int8(p1 - q1);
271
272     a = clip_int8(a);
273
274     // We deviate from the spec here with c(a+3) >> 3
275     // since that's what libvpx does.
276     f1 = FFMIN(a + 4, 127) >> 3;
277
278     if (is_vp7)
279         f2 = f1 - ((a & 7) == 4);
280     else
281         f2 = FFMIN(a + 3, 127) >> 3;
282
283     // Despite what the spec says, we do need to clamp here to
284     // be bitexact with libvpx.
285     p[-1 * stride] = cm[p0 + f2];
286     p[ 0 * stride] = cm[q0 - f1];
287
288     // only used for _inner on blocks without high edge variance
289     if (!is4tap) {
290         a              = (f1 + 1) >> 1;
291         p[-2 * stride] = cm[p1 + a];
292         p[ 1 * stride] = cm[q1 - a];
293     }
294 }
295
296 static av_always_inline void vp7_filter_common(uint8_t *p, ptrdiff_t stride,
297                                                int is4tap)
298 {
299     filter_common(p, stride, is4tap, IS_VP7);
300 }
301
302 static av_always_inline void vp8_filter_common(uint8_t *p, ptrdiff_t stride,
303                                                int is4tap)
304 {
305     filter_common(p, stride, is4tap, IS_VP8);
306 }
307
308 static av_always_inline int vp7_simple_limit(uint8_t *p, ptrdiff_t stride,
309                                              int flim)
310 {
311     LOAD_PIXELS
312     return FFABS(p0 - q0) <= flim;
313 }
314
315 static av_always_inline int vp8_simple_limit(uint8_t *p, ptrdiff_t stride,
316                                              int flim)
317 {
318     LOAD_PIXELS
319     return 2 * FFABS(p0 - q0) + (FFABS(p1 - q1) >> 1) <= flim;
320 }
321
322 /**
323  * E - limit at the macroblock edge
324  * I - limit for interior difference
325  */
326 #define NORMAL_LIMIT(vpn)                                                     \
327 static av_always_inline int vp ## vpn ## _normal_limit(uint8_t *p,            \
328                                                        ptrdiff_t stride,      \
329                                                        int E, int I)          \
330 {                                                                             \
331     LOAD_PIXELS                                                               \
332     return vp ## vpn ## _simple_limit(p, stride, E) &&                        \
333            FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I &&                      \
334            FFABS(p1 - p0) <= I && FFABS(q3 - q2) <= I &&                      \
335            FFABS(q2 - q1) <= I && FFABS(q1 - q0) <= I;                        \
336 }
337
338 NORMAL_LIMIT(7)
339 NORMAL_LIMIT(8)
340
341 // high edge variance
342 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh)
343 {
344     LOAD_PIXELS
345     return FFABS(p1 - p0) > thresh || FFABS(q1 - q0) > thresh;
346 }
347
348 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride)
349 {
350     int a0, a1, a2, w;
351     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
352
353     LOAD_PIXELS
354
355     w = clip_int8(p1 - q1);
356     w = clip_int8(w + 3 * (q0 - p0));
357
358     a0 = (27 * w + 63) >> 7;
359     a1 = (18 * w + 63) >> 7;
360     a2 =  (9 * w + 63) >> 7;
361
362     p[-3 * stride] = cm[p2 + a2];
363     p[-2 * stride] = cm[p1 + a1];
364     p[-1 * stride] = cm[p0 + a0];
365     p[ 0 * stride] = cm[q0 - a0];
366     p[ 1 * stride] = cm[q1 - a1];
367     p[ 2 * stride] = cm[q2 - a2];
368 }
369
370 #define LOOP_FILTER(vpn, dir, size, stridea, strideb, maybe_inline)           \
371 static maybe_inline                                                           \
372 void vpn ## _ ## dir ## _loop_filter ## size ## _c(uint8_t *dst,              \
373                                                    ptrdiff_t stride,          \
374                                                    int flim_E, int flim_I,    \
375                                                    int hev_thresh)            \
376 {                                                                             \
377     int i;                                                                    \
378     for (i = 0; i < size; i++)                                                \
379         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
380                                  flim_E, flim_I)) {                           \
381             if (hev(dst + i * stridea, strideb, hev_thresh))                  \
382                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
383             else                                                              \
384                 filter_mbedge(dst + i * stridea, strideb);                    \
385         }                                                                     \
386 }                                                                             \
387                                                                               \
388 static maybe_inline                                                           \
389 void vpn ## _ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst,        \
390                                                          ptrdiff_t stride,    \
391                                                          int flim_E,          \
392                                                          int flim_I,          \
393                                                          int hev_thresh)      \
394 {                                                                             \
395     int i;                                                                    \
396     for (i = 0; i < size; i++)                                                \
397         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
398                                  flim_E, flim_I)) {                           \
399             int hv = hev(dst + i * stridea, strideb, hev_thresh);             \
400             if (hv)                                                           \
401                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
402             else                                                              \
403                 vpn ## _filter_common(dst + i * stridea, strideb, 0);         \
404         }                                                                     \
405 }
406
407 #define UV_LOOP_FILTER(vpn, dir, stridea, strideb)                            \
408 LOOP_FILTER(vpn, dir, 8, stridea, strideb, av_always_inline)                  \
409 static void vpn ## _ ## dir ## _loop_filter8uv_c(uint8_t *dstU,               \
410                                                  uint8_t *dstV,               \
411                                                  ptrdiff_t stride, int fE,    \
412                                                  int fI, int hev_thresh)      \
413 {                                                                             \
414     vpn ## _ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);     \
415     vpn ## _ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);     \
416 }                                                                             \
417                                                                               \
418 static void vpn ## _ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU,         \
419                                                        uint8_t *dstV,         \
420                                                        ptrdiff_t stride,      \
421                                                        int fE, int fI,        \
422                                                        int hev_thresh)        \
423 {                                                                             \
424     vpn ## _ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI,            \
425                                              hev_thresh);                     \
426     vpn ## _ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI,            \
427                                              hev_thresh);                     \
428 }
429
430 #define LOOP_FILTER_SIMPLE(vpn)                                               \
431 static void vpn ## _v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
432                                            int flim)                          \
433 {                                                                             \
434     int i;                                                                    \
435     for (i = 0; i < 16; i++)                                                  \
436         if (vpn ## _simple_limit(dst + i, stride, flim))                      \
437             vpn ## _filter_common(dst + i, stride, 1);                        \
438 }                                                                             \
439                                                                               \
440 static void vpn ## _h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
441                                            int flim)                          \
442 {                                                                             \
443     int i;                                                                    \
444     for (i = 0; i < 16; i++)                                                  \
445         if (vpn ## _simple_limit(dst + i * stride, 1, flim))                  \
446             vpn ## _filter_common(dst + i * stride, 1, 1);                    \
447 }
448
449 #define LOOP_FILTERS(vpn)                \
450     LOOP_FILTER(vpn, v, 16, 1, stride, ) \
451     LOOP_FILTER(vpn, h, 16, stride, 1, ) \
452     UV_LOOP_FILTER(vpn, v, 1, stride)    \
453     UV_LOOP_FILTER(vpn, h, stride, 1)    \
454     LOOP_FILTER_SIMPLE(vpn)              \
455
456 static const uint8_t subpel_filters[7][6] = {
457     { 0,  6, 123,  12,  1, 0 },
458     { 2, 11, 108,  36,  8, 1 },
459     { 0,  9,  93,  50,  6, 0 },
460     { 3, 16,  77,  77, 16, 3 },
461     { 0,  6,  50,  93,  9, 0 },
462     { 1,  8,  36, 108, 11, 2 },
463     { 0,  1,  12, 123,  6, 0 },
464 };
465
466 #define PUT_PIXELS(WIDTH)                                                     \
467 static void put_vp8_pixels ## WIDTH ## _c(uint8_t *dst, ptrdiff_t dststride,  \
468                                           uint8_t *src, ptrdiff_t srcstride,  \
469                                           int h, int x, int y)                \
470 {                                                                             \
471     int i;                                                                    \
472     for (i = 0; i < h; i++, dst += dststride, src += srcstride)               \
473         memcpy(dst, src, WIDTH);                                              \
474 }
475
476 PUT_PIXELS(16)
477 PUT_PIXELS(8)
478 PUT_PIXELS(4)
479
480 #define FILTER_6TAP(src, F, stride)                                           \
481     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
482         F[0] * src[x - 2 * stride] + F[3] * src[x + 1 * stride] -             \
483         F[4] * src[x + 2 * stride] + F[5] * src[x + 3 * stride] + 64) >> 7]
484
485 #define FILTER_4TAP(src, F, stride)                                           \
486     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
487         F[3] * src[x + 1 * stride] - F[4] * src[x + 2 * stride] + 64) >> 7]
488
489 #define VP8_EPEL_H(SIZE, TAPS)                                                \
490 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst,            \
491                                                      ptrdiff_t dststride,     \
492                                                      uint8_t *src,            \
493                                                      ptrdiff_t srcstride,     \
494                                                      int h, int mx, int my)   \
495 {                                                                             \
496     const uint8_t *filter = subpel_filters[mx - 1];                           \
497     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
498     int x, y;                                                                 \
499     for (y = 0; y < h; y++) {                                                 \
500         for (x = 0; x < SIZE; x++)                                            \
501             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1);                  \
502         dst += dststride;                                                     \
503         src += srcstride;                                                     \
504     }                                                                         \
505 }
506
507 #define VP8_EPEL_V(SIZE, TAPS)                                                \
508 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst,            \
509                                                      ptrdiff_t dststride,     \
510                                                      uint8_t *src,            \
511                                                      ptrdiff_t srcstride,     \
512                                                      int h, int mx, int my)   \
513 {                                                                             \
514     const uint8_t *filter = subpel_filters[my - 1];                           \
515     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
516     int x, y;                                                                 \
517     for (y = 0; y < h; y++) {                                                 \
518         for (x = 0; x < SIZE; x++)                                            \
519             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride);          \
520         dst += dststride;                                                     \
521         src += srcstride;                                                     \
522     }                                                                         \
523 }
524
525 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS)                                       \
526 static void                                                                   \
527 put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst,         \
528                                                         ptrdiff_t dststride,  \
529                                                         uint8_t *src,         \
530                                                         ptrdiff_t srcstride,  \
531                                                         int h, int mx,        \
532                                                         int my)               \
533 {                                                                             \
534     const uint8_t *filter = subpel_filters[mx - 1];                           \
535     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
536     int x, y;                                                                 \
537     uint8_t tmp_array[(2 * SIZE + VTAPS - 1) * SIZE];                         \
538     uint8_t *tmp = tmp_array;                                                 \
539     src -= (2 - (VTAPS == 4)) * srcstride;                                    \
540                                                                               \
541     for (y = 0; y < h + VTAPS - 1; y++) {                                     \
542         for (x = 0; x < SIZE; x++)                                            \
543             tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1);                 \
544         tmp += SIZE;                                                          \
545         src += srcstride;                                                     \
546     }                                                                         \
547     tmp    = tmp_array + (2 - (VTAPS == 4)) * SIZE;                           \
548     filter = subpel_filters[my - 1];                                          \
549                                                                               \
550     for (y = 0; y < h; y++) {                                                 \
551         for (x = 0; x < SIZE; x++)                                            \
552             dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE);              \
553         dst += dststride;                                                     \
554         tmp += SIZE;                                                          \
555     }                                                                         \
556 }
557
558 VP8_EPEL_H(16, 4)
559 VP8_EPEL_H(8,  4)
560 VP8_EPEL_H(4,  4)
561 VP8_EPEL_H(16, 6)
562 VP8_EPEL_H(8,  6)
563 VP8_EPEL_H(4,  6)
564 VP8_EPEL_V(16, 4)
565 VP8_EPEL_V(8,  4)
566 VP8_EPEL_V(4,  4)
567 VP8_EPEL_V(16, 6)
568 VP8_EPEL_V(8,  6)
569 VP8_EPEL_V(4,  6)
570
571 VP8_EPEL_HV(16, 4, 4)
572 VP8_EPEL_HV(8,  4, 4)
573 VP8_EPEL_HV(4,  4, 4)
574 VP8_EPEL_HV(16, 4, 6)
575 VP8_EPEL_HV(8,  4, 6)
576 VP8_EPEL_HV(4,  4, 6)
577 VP8_EPEL_HV(16, 6, 4)
578 VP8_EPEL_HV(8,  6, 4)
579 VP8_EPEL_HV(4,  6, 4)
580 VP8_EPEL_HV(16, 6, 6)
581 VP8_EPEL_HV(8,  6, 6)
582 VP8_EPEL_HV(4,  6, 6)
583
584 #define VP8_BILINEAR(SIZE)                                                    \
585 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t dstride, \
586                                              uint8_t *src, ptrdiff_t sstride, \
587                                              int h, int mx, int my)           \
588 {                                                                             \
589     int a = 8 - mx, b = mx;                                                   \
590     int x, y;                                                                 \
591     for (y = 0; y < h; y++) {                                                 \
592         for (x = 0; x < SIZE; x++)                                            \
593             dst[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
594         dst += dstride;                                                       \
595         src += sstride;                                                       \
596     }                                                                         \
597 }                                                                             \
598                                                                               \
599 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t dstride, \
600                                              uint8_t *src, ptrdiff_t sstride, \
601                                              int h, int mx, int my)           \
602 {                                                                             \
603     int c = 8 - my, d = my;                                                   \
604     int x, y;                                                                 \
605     for (y = 0; y < h; y++) {                                                 \
606         for (x = 0; x < SIZE; x++)                                            \
607             dst[x] = (c * src[x] + d * src[x + sstride] + 4) >> 3;            \
608         dst += dstride;                                                       \
609         src += sstride;                                                       \
610     }                                                                         \
611 }                                                                             \
612                                                                               \
613 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst,                   \
614                                               ptrdiff_t dstride,              \
615                                               uint8_t *src,                   \
616                                               ptrdiff_t sstride,              \
617                                               int h, int mx, int my)          \
618 {                                                                             \
619     int a = 8 - mx, b = mx;                                                   \
620     int c = 8 - my, d = my;                                                   \
621     int x, y;                                                                 \
622     uint8_t tmp_array[(2 * SIZE + 1) * SIZE];                                 \
623     uint8_t *tmp = tmp_array;                                                 \
624     for (y = 0; y < h + 1; y++) {                                             \
625         for (x = 0; x < SIZE; x++)                                            \
626             tmp[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
627         tmp += SIZE;                                                          \
628         src += sstride;                                                       \
629     }                                                                         \
630     tmp = tmp_array;                                                          \
631     for (y = 0; y < h; y++) {                                                 \
632         for (x = 0; x < SIZE; x++)                                            \
633             dst[x] = (c * tmp[x] + d * tmp[x + SIZE] + 4) >> 3;               \
634         dst += dstride;                                                       \
635         tmp += SIZE;                                                          \
636     }                                                                         \
637 }
638
639 VP8_BILINEAR(16)
640 VP8_BILINEAR(8)
641 VP8_BILINEAR(4)
642
643 #define VP78_MC_FUNC(IDX, SIZE)                                               \
644     dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c;   \
645     dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c;  \
646     dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c;  \
647     dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c;  \
648     dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
649     dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
650     dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c;  \
651     dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
652     dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
653
654 #define VP78_BILINEAR_MC_FUNC(IDX, SIZE)                                      \
655     dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels   ## SIZE ## _c; \
656     dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
657     dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
658     dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
659     dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
660     dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
661     dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
662     dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
663     dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
664
665 av_cold void ff_vp78dsp_init(VP8DSPContext *dsp)
666 {
667     VP78_MC_FUNC(0, 16);
668     VP78_MC_FUNC(1, 8);
669     VP78_MC_FUNC(2, 4);
670
671     VP78_BILINEAR_MC_FUNC(0, 16);
672     VP78_BILINEAR_MC_FUNC(1, 8);
673     VP78_BILINEAR_MC_FUNC(2, 4);
674
675     if (ARCH_ARM)
676         ff_vp78dsp_init_arm(dsp);
677     if (ARCH_PPC)
678         ff_vp78dsp_init_ppc(dsp);
679     if (ARCH_X86)
680         ff_vp78dsp_init_x86(dsp);
681 }
682
683 #if CONFIG_VP7_DECODER
684 LOOP_FILTERS(vp7)
685
686 av_cold void ff_vp7dsp_init(VP8DSPContext *dsp)
687 {
688     dsp->vp8_luma_dc_wht    = vp7_luma_dc_wht_c;
689     dsp->vp8_luma_dc_wht_dc = vp7_luma_dc_wht_dc_c;
690     dsp->vp8_idct_add       = vp7_idct_add_c;
691     dsp->vp8_idct_dc_add    = vp7_idct_dc_add_c;
692     dsp->vp8_idct_dc_add4y  = vp7_idct_dc_add4y_c;
693     dsp->vp8_idct_dc_add4uv = vp7_idct_dc_add4uv_c;
694
695     dsp->vp8_v_loop_filter16y = vp7_v_loop_filter16_c;
696     dsp->vp8_h_loop_filter16y = vp7_h_loop_filter16_c;
697     dsp->vp8_v_loop_filter8uv = vp7_v_loop_filter8uv_c;
698     dsp->vp8_h_loop_filter8uv = vp7_h_loop_filter8uv_c;
699
700     dsp->vp8_v_loop_filter16y_inner = vp7_v_loop_filter16_inner_c;
701     dsp->vp8_h_loop_filter16y_inner = vp7_h_loop_filter16_inner_c;
702     dsp->vp8_v_loop_filter8uv_inner = vp7_v_loop_filter8uv_inner_c;
703     dsp->vp8_h_loop_filter8uv_inner = vp7_h_loop_filter8uv_inner_c;
704
705     dsp->vp8_v_loop_filter_simple = vp7_v_loop_filter_simple_c;
706     dsp->vp8_h_loop_filter_simple = vp7_h_loop_filter_simple_c;
707 }
708 #endif /* CONFIG_VP7_DECODER */
709
710 #if CONFIG_VP8_DECODER
711 LOOP_FILTERS(vp8)
712
713 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
714 {
715     dsp->vp8_luma_dc_wht    = vp8_luma_dc_wht_c;
716     dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
717     dsp->vp8_idct_add       = vp8_idct_add_c;
718     dsp->vp8_idct_dc_add    = vp8_idct_dc_add_c;
719     dsp->vp8_idct_dc_add4y  = vp8_idct_dc_add4y_c;
720     dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
721
722     dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
723     dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
724     dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
725     dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
726
727     dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
728     dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
729     dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
730     dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
731
732     dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
733     dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
734
735     if (ARCH_ARM)
736         ff_vp8dsp_init_arm(dsp);
737     if (ARCH_X86)
738         ff_vp8dsp_init_x86(dsp);
739     if (ARCH_MIPS)
740         ff_vp8dsp_init_mips(dsp);
741 }
742 #endif /* CONFIG_VP8_DECODER */