Fix linking if MMX is disabled.
[libav.git] / libavcodec / vp8dsp.c
1 /**
2 * VP8 compatible video decoder
3 *
4 * Copyright (C) 2010 David Conrad
5 * Copyright (C) 2010 Ronald S. Bultje
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 "dsputil.h"
25 #include "vp8dsp.h"
26
27 // TODO: Maybe add dequant
28 static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
29 {
30 int i, t0, t1, t2, t3;
31
32 for (i = 0; i < 4; i++) {
33 t0 = dc[0*4+i] + dc[3*4+i];
34 t1 = dc[1*4+i] + dc[2*4+i];
35 t2 = dc[1*4+i] - dc[2*4+i];
36 t3 = dc[0*4+i] - dc[3*4+i];
37
38 dc[0*4+i] = t0 + t1;
39 dc[1*4+i] = t3 + t2;
40 dc[2*4+i] = t0 - t1;
41 dc[3*4+i] = t3 - t2;
42 }
43
44 for (i = 0; i < 4; i++) {
45 t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding
46 t1 = dc[i*4+1] + dc[i*4+2];
47 t2 = dc[i*4+1] - dc[i*4+2];
48 t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
49
50 *block[i][0] = (t0 + t1) >> 3;
51 *block[i][1] = (t3 + t2) >> 3;
52 *block[i][2] = (t0 - t1) >> 3;
53 *block[i][3] = (t3 - t2) >> 3;
54 }
55 }
56
57
58 #define MUL_20091(a) ((((a)*20091) >> 16) + (a))
59 #define MUL_35468(a) (((a)*35468) >> 16)
60
61 static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
62 {
63 int i, t0, t1, t2, t3;
64 DCTELEM tmp[16];
65
66 for (i = 0; i < 4; i++) {
67 t0 = block[0*4+i] + block[2*4+i];
68 t1 = block[0*4+i] - block[2*4+i];
69 t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
70 t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
71
72 tmp[i*4+0] = t0 + t3;
73 tmp[i*4+1] = t1 + t2;
74 tmp[i*4+2] = t1 - t2;
75 tmp[i*4+3] = t0 - t3;
76 }
77
78 for (i = 0; i < 4; i++) {
79 t0 = tmp[0*4+i] + tmp[2*4+i];
80 t1 = tmp[0*4+i] - tmp[2*4+i];
81 t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
82 t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
83
84 dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
85 dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
86 dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
87 dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
88 dst += stride;
89 }
90 }
91
92 static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
93 {
94 int i, dc = (block[0] + 4) >> 3;
95
96 for (i = 0; i < 4; i++) {
97 dst[0] = av_clip_uint8(dst[0] + dc);
98 dst[1] = av_clip_uint8(dst[1] + dc);
99 dst[2] = av_clip_uint8(dst[2] + dc);
100 dst[3] = av_clip_uint8(dst[3] + dc);
101 dst += stride;
102 }
103 }
104
105
106 // because I like only having two parameters to pass functions...
107 #define LOAD_PIXELS\
108 int av_unused p3 = p[-4*stride];\
109 int av_unused p2 = p[-3*stride];\
110 int av_unused p1 = p[-2*stride];\
111 int av_unused p0 = p[-1*stride];\
112 int av_unused q0 = p[ 0*stride];\
113 int av_unused q1 = p[ 1*stride];\
114 int av_unused q2 = p[ 2*stride];\
115 int av_unused q3 = p[ 3*stride];
116
117 static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
118 {
119 LOAD_PIXELS
120 int a, f1, f2;
121
122 a = 3*(q0 - p0);
123
124 if (is4tap)
125 a += av_clip_int8(p1 - q1);
126
127 a = av_clip_int8(a);
128
129 // We deviate from the spec here with c(a+3) >> 3
130 // since that's what libvpx does.
131 f1 = FFMIN(a+4, 127) >> 3;
132 f2 = FFMIN(a+3, 127) >> 3;
133
134 // Despite what the spec says, we do need to clamp here to
135 // be bitexact with libvpx.
136 p[-1*stride] = av_clip_uint8(p0 + f2);
137 p[ 0*stride] = av_clip_uint8(q0 - f1);
138
139 // only used for _inner on blocks without high edge variance
140 if (!is4tap) {
141 a = (f1+1)>>1;
142 p[-2*stride] = av_clip_uint8(p1 + a);
143 p[ 1*stride] = av_clip_uint8(q1 - a);
144 }
145 }
146
147 static av_always_inline int simple_limit(uint8_t *p, int stride, int flim)
148 {
149 LOAD_PIXELS
150 return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
151 }
152
153 /**
154 * E - limit at the macroblock edge
155 * I - limit for interior difference
156 */
157 static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I)
158 {
159 LOAD_PIXELS
160 return simple_limit(p, stride, 2*E+I)
161 && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
162 && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
163 }
164
165 // high edge variance
166 static av_always_inline int hev(uint8_t *p, int stride, int thresh)
167 {
168 LOAD_PIXELS
169 return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
170 }
171
172 static av_always_inline void filter_mbedge(uint8_t *p, int stride)
173 {
174 int a0, a1, a2, w;
175
176 LOAD_PIXELS
177
178 w = av_clip_int8(p1-q1);
179 w = av_clip_int8(w + 3*(q0-p0));
180
181 a0 = (27*w + 63) >> 7;
182 a1 = (18*w + 63) >> 7;
183 a2 = ( 9*w + 63) >> 7;
184
185 p[-3*stride] = av_clip_uint8(p2 + a2);
186 p[-2*stride] = av_clip_uint8(p1 + a1);
187 p[-1*stride] = av_clip_uint8(p0 + a0);
188 p[ 0*stride] = av_clip_uint8(q0 - a0);
189 p[ 1*stride] = av_clip_uint8(q1 - a1);
190 p[ 2*stride] = av_clip_uint8(q2 - a2);
191 }
192
193 #define LOOP_FILTER(dir, size, stridea, strideb) \
194 static void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
195 int flim_E, int flim_I, int hev_thresh)\
196 {\
197 int i;\
198 \
199 for (i = 0; i < size; i++)\
200 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
201 if (hev(dst+i*stridea, strideb, hev_thresh))\
202 filter_common(dst+i*stridea, strideb, 1);\
203 else\
204 filter_mbedge(dst+i*stridea, strideb);\
205 }\
206 }\
207 \
208 static void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
209 int flim_E, int flim_I, int hev_thresh)\
210 {\
211 int i, hv;\
212 \
213 for (i = 0; i < size; i++)\
214 if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
215 hv = hev(dst+i*stridea, strideb, hev_thresh);\
216 filter_common(dst+i*stridea, strideb, hv);\
217 }\
218 }
219
220 LOOP_FILTER(v, 16, 1, stride)
221 LOOP_FILTER(h, 16, stride, 1)
222 LOOP_FILTER(v, 8, 1, stride)
223 LOOP_FILTER(h, 8, stride, 1)
224
225 static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
226 {
227 int i;
228
229 for (i = 0; i < 16; i++)
230 if (simple_limit(dst+i, stride, flim))
231 filter_common(dst+i, stride, 1);
232 }
233
234 static void vp8_h_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
235 {
236 int i;
237
238 for (i = 0; i < 16; i++)
239 if (simple_limit(dst+i*stride, 1, flim))
240 filter_common(dst+i*stride, 1, 1);
241 }
242
243 static const uint8_t subpel_filters[7][6] = {
244 { 0, 6, 123, 12, 1, 0 },
245 { 2, 11, 108, 36, 8, 1 },
246 { 0, 9, 93, 50, 6, 0 },
247 { 3, 16, 77, 77, 16, 3 },
248 { 0, 6, 50, 93, 9, 0 },
249 { 1, 8, 36, 108, 11, 2 },
250 { 0, 1, 12, 123, 6, 0 },
251 };
252
253 #define PUT_PIXELS(WIDTH) \
254 static void put_vp8_pixels ## WIDTH ##_c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int x, int y) { \
255 int i; \
256 for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \
257 memcpy(dst, src, WIDTH); \
258 } \
259 }
260
261 PUT_PIXELS(16)
262 PUT_PIXELS(8)
263 PUT_PIXELS(4)
264
265 #define FILTER_6TAP(src, F, stride) \
266 av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
267 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7)
268
269 #define FILTER_4TAP(src, F, stride) \
270 av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
271 F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7)
272
273 #define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \
274 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
275 { \
276 const uint8_t *filter = subpel_filters[mx-1]; \
277 int x, y; \
278 \
279 for (y = 0; y < h; y++) { \
280 for (x = 0; x < SIZE; x++) \
281 dst[x] = FILTER(src, filter, 1); \
282 dst += dststride; \
283 src += srcstride; \
284 } \
285 }
286 #define VP8_EPEL_V(SIZE, FILTER, FILTERNAME) \
287 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
288 { \
289 const uint8_t *filter = subpel_filters[my-1]; \
290 int x, y; \
291 \
292 for (y = 0; y < h; y++) { \
293 for (x = 0; x < SIZE; x++) \
294 dst[x] = FILTER(src, filter, srcstride); \
295 dst += dststride; \
296 src += srcstride; \
297 } \
298 }
299 #define VP8_EPEL_HV(SIZE, FILTERX, FILTERY, FILTERNAME) \
300 static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
301 { \
302 const uint8_t *filter = subpel_filters[mx-1]; \
303 int x, y; \
304 uint8_t tmp_array[(2*SIZE+5)*SIZE]; \
305 uint8_t *tmp = tmp_array; \
306 src -= 2*srcstride; \
307 \
308 for (y = 0; y < h+5; y++) { \
309 for (x = 0; x < SIZE; x++) \
310 tmp[x] = FILTERX(src, filter, 1); \
311 tmp += SIZE; \
312 src += srcstride; \
313 } \
314 \
315 tmp = tmp_array + 2*SIZE; \
316 filter = subpel_filters[my-1]; \
317 \
318 for (y = 0; y < h; y++) { \
319 for (x = 0; x < SIZE; x++) \
320 dst[x] = FILTERY(tmp, filter, SIZE); \
321 dst += dststride; \
322 tmp += SIZE; \
323 } \
324 }
325
326 VP8_EPEL_H(16, FILTER_4TAP, h4)
327 VP8_EPEL_H(8, FILTER_4TAP, h4)
328 VP8_EPEL_H(4, FILTER_4TAP, h4)
329 VP8_EPEL_H(16, FILTER_6TAP, h6)
330 VP8_EPEL_H(8, FILTER_6TAP, h6)
331 VP8_EPEL_H(4, FILTER_6TAP, h6)
332 VP8_EPEL_V(16, FILTER_4TAP, v4)
333 VP8_EPEL_V(8, FILTER_4TAP, v4)
334 VP8_EPEL_V(4, FILTER_4TAP, v4)
335 VP8_EPEL_V(16, FILTER_6TAP, v6)
336 VP8_EPEL_V(8, FILTER_6TAP, v6)
337 VP8_EPEL_V(4, FILTER_6TAP, v6)
338 VP8_EPEL_HV(16, FILTER_4TAP, FILTER_4TAP, h4v4)
339 VP8_EPEL_HV(8, FILTER_4TAP, FILTER_4TAP, h4v4)
340 VP8_EPEL_HV(4, FILTER_4TAP, FILTER_4TAP, h4v4)
341 VP8_EPEL_HV(16, FILTER_4TAP, FILTER_6TAP, h4v6)
342 VP8_EPEL_HV(8, FILTER_4TAP, FILTER_6TAP, h4v6)
343 VP8_EPEL_HV(4, FILTER_4TAP, FILTER_6TAP, h4v6)
344 VP8_EPEL_HV(16, FILTER_6TAP, FILTER_4TAP, h6v4)
345 VP8_EPEL_HV(8, FILTER_6TAP, FILTER_4TAP, h6v4)
346 VP8_EPEL_HV(4, FILTER_6TAP, FILTER_4TAP, h6v4)
347 VP8_EPEL_HV(16, FILTER_6TAP, FILTER_6TAP, h6v6)
348 VP8_EPEL_HV(8, FILTER_6TAP, FILTER_6TAP, h6v6)
349 VP8_EPEL_HV(4, FILTER_6TAP, FILTER_6TAP, h6v6)
350
351 #define VP8_BILINEAR(SIZE) \
352 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
353 { \
354 int a = 8-mx, b = mx; \
355 int x, y; \
356 \
357 for (y = 0; y < h; y++) { \
358 for (x = 0; x < SIZE; x++) \
359 dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
360 dst += stride; \
361 src += stride; \
362 } \
363 } \
364 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
365 { \
366 int c = 8-my, d = my; \
367 int x, y; \
368 \
369 for (y = 0; y < h; y++) { \
370 for (x = 0; x < SIZE; x++) \
371 dst[x] = (c*src[x] + d*src[x+stride] + 4) >> 3; \
372 dst += stride; \
373 src += stride; \
374 } \
375 } \
376 \
377 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
378 { \
379 int a = 8-mx, b = mx; \
380 int c = 8-my, d = my; \
381 int x, y; \
382 uint8_t tmp_array[(2*SIZE+1)*SIZE]; \
383 uint8_t *tmp = tmp_array; \
384 \
385 for (y = 0; y < h+1; y++) { \
386 for (x = 0; x < SIZE; x++) \
387 tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
388 tmp += SIZE; \
389 src += stride; \
390 } \
391 \
392 tmp = tmp_array; \
393 \
394 for (y = 0; y < h; y++) { \
395 for (x = 0; x < SIZE; x++) \
396 dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \
397 dst += stride; \
398 tmp += SIZE; \
399 } \
400 }
401
402 VP8_BILINEAR(16)
403 VP8_BILINEAR(8)
404 VP8_BILINEAR(4)
405
406 #define VP8_MC_FUNC(IDX, SIZE) \
407 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
408 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
409 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
410 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
411 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
412 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
413 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
414 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
415 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
416
417 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \
418 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
419 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
420 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
421 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
422 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
423 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
424 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
425 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
426 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
427
428 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
429 {
430 dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
431 dsp->vp8_idct_add = vp8_idct_add_c;
432 dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
433
434 dsp->vp8_v_loop_filter16 = vp8_v_loop_filter16_c;
435 dsp->vp8_h_loop_filter16 = vp8_h_loop_filter16_c;
436 dsp->vp8_v_loop_filter8 = vp8_v_loop_filter8_c;
437 dsp->vp8_h_loop_filter8 = vp8_h_loop_filter8_c;
438
439 dsp->vp8_v_loop_filter16_inner = vp8_v_loop_filter16_inner_c;
440 dsp->vp8_h_loop_filter16_inner = vp8_h_loop_filter16_inner_c;
441 dsp->vp8_v_loop_filter8_inner = vp8_v_loop_filter8_inner_c;
442 dsp->vp8_h_loop_filter8_inner = vp8_h_loop_filter8_inner_c;
443
444 dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
445 dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
446
447 VP8_MC_FUNC(0, 16);
448 VP8_MC_FUNC(1, 8);
449 VP8_MC_FUNC(2, 4);
450
451 VP8_BILINEAR_MC_FUNC(0, 16);
452 VP8_BILINEAR_MC_FUNC(1, 8);
453 VP8_BILINEAR_MC_FUNC(2, 4);
454
455 if (HAVE_MMX)
456 ff_vp8dsp_init_x86(dsp);
457 }