vp8dsp.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2010 David Conrad
3  * Copyright (C) 2010 Ronald S. Bultje
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
27 #include "dsputil.h"
28 #include "vp8dsp.h"
29 
30 // TODO: Maybe add dequant
31 static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
32 {
33  int i, t0, t1, t2, t3;
34 
35  for (i = 0; i < 4; i++) {
36  t0 = dc[0*4+i] + dc[3*4+i];
37  t1 = dc[1*4+i] + dc[2*4+i];
38  t2 = dc[1*4+i] - dc[2*4+i];
39  t3 = dc[0*4+i] - dc[3*4+i];
40 
41  dc[0*4+i] = t0 + t1;
42  dc[1*4+i] = t3 + t2;
43  dc[2*4+i] = t0 - t1;
44  dc[3*4+i] = t3 - t2;
45  }
46 
47  for (i = 0; i < 4; i++) {
48  t0 = dc[i*4+0] + dc[i*4+3] + 3; // rounding
49  t1 = dc[i*4+1] + dc[i*4+2];
50  t2 = dc[i*4+1] - dc[i*4+2];
51  t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
52  dc[i*4+0] = 0;
53  dc[i*4+1] = 0;
54  dc[i*4+2] = 0;
55  dc[i*4+3] = 0;
56 
57  block[i][0][0] = (t0 + t1) >> 3;
58  block[i][1][0] = (t3 + t2) >> 3;
59  block[i][2][0] = (t0 - t1) >> 3;
60  block[i][3][0] = (t3 - t2) >> 3;
61  }
62 }
63 
64 static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16])
65 {
66  int i, val = (dc[0] + 3) >> 3;
67  dc[0] = 0;
68 
69  for (i = 0; i < 4; i++) {
70  block[i][0][0] = val;
71  block[i][1][0] = val;
72  block[i][2][0] = val;
73  block[i][3][0] = val;
74  }
75 }
76 
77 #define MUL_20091(a) ((((a)*20091) >> 16) + (a))
78 #define MUL_35468(a) (((a)*35468) >> 16)
79 
80 static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
81 {
82  int i, t0, t1, t2, t3;
83  DCTELEM tmp[16];
84 
85  for (i = 0; i < 4; i++) {
86  t0 = block[0*4+i] + block[2*4+i];
87  t1 = block[0*4+i] - block[2*4+i];
88  t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
89  t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
90  block[0*4+i] = 0;
91  block[1*4+i] = 0;
92  block[2*4+i] = 0;
93  block[3*4+i] = 0;
94 
95  tmp[i*4+0] = t0 + t3;
96  tmp[i*4+1] = t1 + t2;
97  tmp[i*4+2] = t1 - t2;
98  tmp[i*4+3] = t0 - t3;
99  }
100 
101  for (i = 0; i < 4; i++) {
102  t0 = tmp[0*4+i] + tmp[2*4+i];
103  t1 = tmp[0*4+i] - tmp[2*4+i];
104  t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
105  t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
106 
107  dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
108  dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
109  dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
110  dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
111  dst += stride;
112  }
113 }
114 
115 static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
116 {
117  int i, dc = (block[0] + 4) >> 3;
118  block[0] = 0;
119 
120  for (i = 0; i < 4; i++) {
121  dst[0] = av_clip_uint8(dst[0] + dc);
122  dst[1] = av_clip_uint8(dst[1] + dc);
123  dst[2] = av_clip_uint8(dst[2] + dc);
124  dst[3] = av_clip_uint8(dst[3] + dc);
125  dst += stride;
126  }
127 }
128 
129 static void vp8_idct_dc_add4uv_c(uint8_t *dst, DCTELEM block[4][16], int stride)
130 {
131  vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride);
132  vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride);
133  vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride);
134  vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride);
135 }
136 
137 static void vp8_idct_dc_add4y_c(uint8_t *dst, DCTELEM block[4][16], int stride)
138 {
139  vp8_idct_dc_add_c(dst+ 0, block[0], stride);
140  vp8_idct_dc_add_c(dst+ 4, block[1], stride);
141  vp8_idct_dc_add_c(dst+ 8, block[2], stride);
142  vp8_idct_dc_add_c(dst+12, block[3], stride);
143 }
144 
145 // because I like only having two parameters to pass functions...
146 #define LOAD_PIXELS\
147  int av_unused p3 = p[-4*stride];\
148  int av_unused p2 = p[-3*stride];\
149  int av_unused p1 = p[-2*stride];\
150  int av_unused p0 = p[-1*stride];\
151  int av_unused q0 = p[ 0*stride];\
152  int av_unused q1 = p[ 1*stride];\
153  int av_unused q2 = p[ 2*stride];\
154  int av_unused q3 = p[ 3*stride];
155 
156 #define clip_int8(n) (cm[n+0x80]-0x80)
157 
158 static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
159 {
161  int a, f1, f2;
162  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
163 
164  a = 3*(q0 - p0);
165 
166  if (is4tap)
167  a += clip_int8(p1 - q1);
168 
169  a = clip_int8(a);
170 
171  // We deviate from the spec here with c(a+3) >> 3
172  // since that's what libvpx does.
173  f1 = FFMIN(a+4, 127) >> 3;
174  f2 = FFMIN(a+3, 127) >> 3;
175 
176  // Despite what the spec says, we do need to clamp here to
177  // be bitexact with libvpx.
178  p[-1*stride] = cm[p0 + f2];
179  p[ 0*stride] = cm[q0 - f1];
180 
181  // only used for _inner on blocks without high edge variance
182  if (!is4tap) {
183  a = (f1+1)>>1;
184  p[-2*stride] = cm[p1 + a];
185  p[ 1*stride] = cm[q1 - a];
186  }
187 }
188 
189 static av_always_inline int simple_limit(uint8_t *p, int stride, int flim)
190 {
192  return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
193 }
194 
199 static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I)
200 {
202  return simple_limit(p, stride, E)
203  && FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
204  && FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
205 }
206 
207 // high edge variance
208 static av_always_inline int hev(uint8_t *p, int stride, int thresh)
209 {
211  return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
212 }
213 
214 static av_always_inline void filter_mbedge(uint8_t *p, int stride)
215 {
216  int a0, a1, a2, w;
217  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
218 
220 
221  w = clip_int8(p1-q1);
222  w = clip_int8(w + 3*(q0-p0));
223 
224  a0 = (27*w + 63) >> 7;
225  a1 = (18*w + 63) >> 7;
226  a2 = ( 9*w + 63) >> 7;
227 
228  p[-3*stride] = cm[p2 + a2];
229  p[-2*stride] = cm[p1 + a1];
230  p[-1*stride] = cm[p0 + a0];
231  p[ 0*stride] = cm[q0 - a0];
232  p[ 1*stride] = cm[q1 - a1];
233  p[ 2*stride] = cm[q2 - a2];
234 }
235 
236 #define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
237 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
238  int flim_E, int flim_I, int hev_thresh)\
239 {\
240  int i;\
241 \
242  for (i = 0; i < size; i++)\
243  if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
244  if (hev(dst+i*stridea, strideb, hev_thresh))\
245  filter_common(dst+i*stridea, strideb, 1);\
246  else\
247  filter_mbedge(dst+i*stridea, strideb);\
248  }\
249 }\
250 \
251 static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
252  int flim_E, int flim_I, int hev_thresh)\
253 {\
254  int i;\
255 \
256  for (i = 0; i < size; i++)\
257  if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
258  int hv = hev(dst+i*stridea, strideb, hev_thresh);\
259  if (hv) \
260  filter_common(dst+i*stridea, strideb, 1);\
261  else \
262  filter_common(dst+i*stridea, strideb, 0);\
263  }\
264 }
265 
266 LOOP_FILTER(v, 16, 1, stride,)
267 LOOP_FILTER(h, 16, stride, 1,)
268 
269 #define UV_LOOP_FILTER(dir, stridea, strideb) \
270 LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
271 static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, int stride,\
272  int fE, int fI, int hev_thresh)\
273 {\
274  vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
275  vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
276 }\
277 static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, int stride,\
278  int fE, int fI, int hev_thresh)\
279 {\
280  vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
281  vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
282 }
283 
284 UV_LOOP_FILTER(v, 1, stride)
285 UV_LOOP_FILTER(h, stride, 1)
286 
287 static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
288 {
289  int i;
290 
291  for (i = 0; i < 16; i++)
292  if (simple_limit(dst+i, stride, flim))
293  filter_common(dst+i, stride, 1);
294 }
295 
296 static void vp8_h_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
297 {
298  int i;
299 
300  for (i = 0; i < 16; i++)
301  if (simple_limit(dst+i*stride, 1, flim))
302  filter_common(dst+i*stride, 1, 1);
303 }
304 
305 static const uint8_t subpel_filters[7][6] = {
306  { 0, 6, 123, 12, 1, 0 },
307  { 2, 11, 108, 36, 8, 1 },
308  { 0, 9, 93, 50, 6, 0 },
309  { 3, 16, 77, 77, 16, 3 },
310  { 0, 6, 50, 93, 9, 0 },
311  { 1, 8, 36, 108, 11, 2 },
312  { 0, 1, 12, 123, 6, 0 },
313 };
314 
315 #define PUT_PIXELS(WIDTH) \
316 static void put_vp8_pixels ## WIDTH ##_c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int x, int y) { \
317  int i; \
318  for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \
319  memcpy(dst, src, WIDTH); \
320  } \
321 }
322 
323 PUT_PIXELS(16)
324 PUT_PIXELS(8)
325 PUT_PIXELS(4)
326 
327 #define FILTER_6TAP(src, F, stride) \
328  cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
329  F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
330 
331 #define FILTER_4TAP(src, F, stride) \
332  cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
333  F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
334 
335 #define VP8_EPEL_H(SIZE, TAPS) \
336 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
337 { \
338  const uint8_t *filter = subpel_filters[mx-1]; \
339  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
340  int x, y; \
341 \
342  for (y = 0; y < h; y++) { \
343  for (x = 0; x < SIZE; x++) \
344  dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1); \
345  dst += dststride; \
346  src += srcstride; \
347  } \
348 }
349 #define VP8_EPEL_V(SIZE, TAPS) \
350 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
351 { \
352  const uint8_t *filter = subpel_filters[my-1]; \
353  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
354  int x, y; \
355 \
356  for (y = 0; y < h; y++) { \
357  for (x = 0; x < SIZE; x++) \
358  dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride); \
359  dst += dststride; \
360  src += srcstride; \
361  } \
362 }
363 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS) \
364 static void put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
365 { \
366  const uint8_t *filter = subpel_filters[mx-1]; \
367  uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
368  int x, y; \
369  uint8_t tmp_array[(2*SIZE+VTAPS-1)*SIZE]; \
370  uint8_t *tmp = tmp_array; \
371  src -= (2-(VTAPS==4))*srcstride; \
372 \
373  for (y = 0; y < h+VTAPS-1; y++) { \
374  for (x = 0; x < SIZE; x++) \
375  tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1); \
376  tmp += SIZE; \
377  src += srcstride; \
378  } \
379 \
380  tmp = tmp_array + (2-(VTAPS==4))*SIZE; \
381  filter = subpel_filters[my-1]; \
382 \
383  for (y = 0; y < h; y++) { \
384  for (x = 0; x < SIZE; x++) \
385  dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE); \
386  dst += dststride; \
387  tmp += SIZE; \
388  } \
389 }
390 
391 VP8_EPEL_H(16, 4)
392 VP8_EPEL_H(8, 4)
393 VP8_EPEL_H(4, 4)
394 VP8_EPEL_H(16, 6)
395 VP8_EPEL_H(8, 6)
396 VP8_EPEL_H(4, 6)
397 VP8_EPEL_V(16, 4)
398 VP8_EPEL_V(8, 4)
399 VP8_EPEL_V(4, 4)
400 VP8_EPEL_V(16, 6)
401 VP8_EPEL_V(8, 6)
402 VP8_EPEL_V(4, 6)
403 VP8_EPEL_HV(16, 4, 4)
404 VP8_EPEL_HV(8, 4, 4)
405 VP8_EPEL_HV(4, 4, 4)
406 VP8_EPEL_HV(16, 4, 6)
407 VP8_EPEL_HV(8, 4, 6)
408 VP8_EPEL_HV(4, 4, 6)
409 VP8_EPEL_HV(16, 6, 4)
410 VP8_EPEL_HV(8, 6, 4)
411 VP8_EPEL_HV(4, 6, 4)
412 VP8_EPEL_HV(16, 6, 6)
413 VP8_EPEL_HV(8, 6, 6)
414 VP8_EPEL_HV(4, 6, 6)
415 
416 #define VP8_BILINEAR(SIZE) \
417 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
418 { \
419  int a = 8-mx, b = mx; \
420  int x, y; \
421 \
422  for (y = 0; y < h; y++) { \
423  for (x = 0; x < SIZE; x++) \
424  dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
425  dst += stride; \
426  src += stride; \
427  } \
428 } \
429 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
430 { \
431  int c = 8-my, d = my; \
432  int x, y; \
433 \
434  for (y = 0; y < h; y++) { \
435  for (x = 0; x < SIZE; x++) \
436  dst[x] = (c*src[x] + d*src[x+stride] + 4) >> 3; \
437  dst += stride; \
438  src += stride; \
439  } \
440 } \
441 \
442 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, int stride, uint8_t *src, int s2, int h, int mx, int my) \
443 { \
444  int a = 8-mx, b = mx; \
445  int c = 8-my, d = my; \
446  int x, y; \
447  uint8_t tmp_array[(2*SIZE+1)*SIZE]; \
448  uint8_t *tmp = tmp_array; \
449 \
450  for (y = 0; y < h+1; y++) { \
451  for (x = 0; x < SIZE; x++) \
452  tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
453  tmp += SIZE; \
454  src += stride; \
455  } \
456 \
457  tmp = tmp_array; \
458 \
459  for (y = 0; y < h; y++) { \
460  for (x = 0; x < SIZE; x++) \
461  dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \
462  dst += stride; \
463  tmp += SIZE; \
464  } \
465 }
466 
467 VP8_BILINEAR(16)
468 VP8_BILINEAR(8)
469 VP8_BILINEAR(4)
470 
471 #define VP8_MC_FUNC(IDX, SIZE) \
472  dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
473  dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
474  dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
475  dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
476  dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
477  dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
478  dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
479  dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
480  dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
481 
482 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \
483  dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
484  dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
485  dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
486  dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
487  dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
488  dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
489  dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
490  dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
491  dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
492 
494 {
501 
502  dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
503  dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
504  dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
505  dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
506 
507  dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
508  dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
509  dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
510  dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
511 
514 
515  VP8_MC_FUNC(0, 16);
516  VP8_MC_FUNC(1, 8);
517  VP8_MC_FUNC(2, 4);
518 
519  VP8_BILINEAR_MC_FUNC(0, 16);
520  VP8_BILINEAR_MC_FUNC(1, 8);
521  VP8_BILINEAR_MC_FUNC(2, 4);
522 
523  if (HAVE_MMX)
524  ff_vp8dsp_init_x86(dsp);
525  if (HAVE_ALTIVEC)
527  if (ARCH_ARM)
528  ff_vp8dsp_init_arm(dsp);
529 }