2
* VP8 compatible video decoder
4
* Copyright (C) 2010 David Conrad
5
* Copyright (C) 2010 Ronald S. Bultje
7
* This file is part of FFmpeg.
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.
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.
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
27
// TODO: Maybe add dequant
28
static void vp8_luma_dc_wht_c(DCTELEM block[4][4][16], DCTELEM dc[16])
30
int i, t0, t1, t2, t3;
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];
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
54
block[i][0][0] = (t0 + t1) >> 3;
55
block[i][1][0] = (t3 + t2) >> 3;
56
block[i][2][0] = (t0 - t1) >> 3;
57
block[i][3][0] = (t3 - t2) >> 3;
61
static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16])
63
int i, val = (dc[0] + 3) >> 3;
66
for (i = 0; i < 4; i++) {
74
#define MUL_20091(a) ((((a)*20091) >> 16) + (a))
75
#define MUL_35468(a) (((a)*35468) >> 16)
77
static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
79
int i, t0, t1, t2, t3;
80
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
83
for (i = 0; i < 4; i++) {
84
t0 = block[0*4+i] + block[2*4+i];
85
t1 = block[0*4+i] - block[2*4+i];
86
t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
87
t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
99
for (i = 0; i < 4; i++) {
100
t0 = tmp[0*4+i] + tmp[2*4+i];
101
t1 = tmp[0*4+i] - tmp[2*4+i];
102
t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
103
t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
105
dst[0] = cm[dst[0] + ((t0 + t3 + 4) >> 3)];
106
dst[1] = cm[dst[1] + ((t1 + t2 + 4) >> 3)];
107
dst[2] = cm[dst[2] + ((t1 - t2 + 4) >> 3)];
108
dst[3] = cm[dst[3] + ((t0 - t3 + 4) >> 3)];
113
static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
115
int i, dc = (block[0] + 4) >> 3;
116
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP + dc;
119
for (i = 0; i < 4; i++) {
128
static void vp8_idct_dc_add4uv_c(uint8_t *dst, DCTELEM block[4][16], int stride)
130
vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride);
131
vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride);
132
vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride);
133
vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride);
136
static void vp8_idct_dc_add4y_c(uint8_t *dst, DCTELEM block[4][16], int stride)
138
vp8_idct_dc_add_c(dst+ 0, block[0], stride);
139
vp8_idct_dc_add_c(dst+ 4, block[1], stride);
140
vp8_idct_dc_add_c(dst+ 8, block[2], stride);
141
vp8_idct_dc_add_c(dst+12, block[3], stride);
144
// because I like only having two parameters to pass functions...
146
int av_unused p3 = p[-4*stride];\
147
int av_unused p2 = p[-3*stride];\
148
int av_unused p1 = p[-2*stride];\
149
int av_unused p0 = p[-1*stride];\
150
int av_unused q0 = p[ 0*stride];\
151
int av_unused q1 = p[ 1*stride];\
152
int av_unused q2 = p[ 2*stride];\
153
int av_unused q3 = p[ 3*stride];
155
#define clip_int8(n) (cm[n+0x80]-0x80)
157
static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
161
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
166
a += clip_int8(p1 - q1);
170
// We deviate from the spec here with c(a+3) >> 3
171
// since that's what libvpx does.
172
f1 = FFMIN(a+4, 127) >> 3;
173
f2 = FFMIN(a+3, 127) >> 3;
175
// Despite what the spec says, we do need to clamp here to
176
// be bitexact with libvpx.
177
p[-1*stride] = cm[p0 + f2];
178
p[ 0*stride] = cm[q0 - f1];
180
// only used for _inner on blocks without high edge variance
183
p[-2*stride] = cm[p1 + a];
184
p[ 1*stride] = cm[q1 - a];
188
static av_always_inline int simple_limit(uint8_t *p, int stride, int flim)
191
return 2*FFABS(p0-q0) + (FFABS(p1-q1) >> 1) <= flim;
195
* E - limit at the macroblock edge
196
* I - limit for interior difference
198
static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I)
201
return simple_limit(p, stride, E)
202
&& FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
203
&& FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
206
// high edge variance
207
static av_always_inline int hev(uint8_t *p, int stride, int thresh)
210
return FFABS(p1-p0) > thresh || FFABS(q1-q0) > thresh;
213
static av_always_inline void filter_mbedge(uint8_t *p, int stride)
216
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
220
w = clip_int8(p1-q1);
221
w = clip_int8(w + 3*(q0-p0));
223
a0 = (27*w + 63) >> 7;
224
a1 = (18*w + 63) >> 7;
225
a2 = ( 9*w + 63) >> 7;
227
p[-3*stride] = cm[p2 + a2];
228
p[-2*stride] = cm[p1 + a1];
229
p[-1*stride] = cm[p0 + a0];
230
p[ 0*stride] = cm[q0 - a0];
231
p[ 1*stride] = cm[q1 - a1];
232
p[ 2*stride] = cm[q2 - a2];
235
#define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
236
static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
237
int flim_E, int flim_I, int hev_thresh)\
241
for (i = 0; i < size; i++)\
242
if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
243
if (hev(dst+i*stridea, strideb, hev_thresh))\
244
filter_common(dst+i*stridea, strideb, 1);\
246
filter_mbedge(dst+i*stridea, strideb);\
250
static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
251
int flim_E, int flim_I, int hev_thresh)\
255
for (i = 0; i < size; i++)\
256
if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
257
int hv = hev(dst+i*stridea, strideb, hev_thresh);\
259
filter_common(dst+i*stridea, strideb, 1);\
261
filter_common(dst+i*stridea, strideb, 0);\
265
LOOP_FILTER(v, 16, 1, stride,)
266
LOOP_FILTER(h, 16, stride, 1,)
268
#define UV_LOOP_FILTER(dir, stridea, strideb) \
269
LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
270
static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, int stride,\
271
int fE, int fI, int hev_thresh)\
273
vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
274
vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
276
static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, int stride,\
277
int fE, int fI, int hev_thresh)\
279
vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
280
vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
283
UV_LOOP_FILTER(v, 1, stride)
284
UV_LOOP_FILTER(h, stride, 1)
286
static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
290
for (i = 0; i < 16; i++)
291
if (simple_limit(dst+i, stride, flim))
292
filter_common(dst+i, stride, 1);
295
static void vp8_h_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
299
for (i = 0; i < 16; i++)
300
if (simple_limit(dst+i*stride, 1, flim))
301
filter_common(dst+i*stride, 1, 1);
304
static const uint8_t subpel_filters[7][6] = {
305
{ 0, 6, 123, 12, 1, 0 },
306
{ 2, 11, 108, 36, 8, 1 },
307
{ 0, 9, 93, 50, 6, 0 },
308
{ 3, 16, 77, 77, 16, 3 },
309
{ 0, 6, 50, 93, 9, 0 },
310
{ 1, 8, 36, 108, 11, 2 },
311
{ 0, 1, 12, 123, 6, 0 },
314
#define PUT_PIXELS(WIDTH) \
315
static void put_vp8_pixels ## WIDTH ##_c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int x, int y) { \
317
for (i = 0; i < h; i++, dst+= dststride, src+= srcstride) { \
318
memcpy(dst, src, WIDTH); \
326
#define FILTER_6TAP(src, F, stride) \
327
cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
328
F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
330
#define FILTER_4TAP(src, F, stride) \
331
cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
332
F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
334
#define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \
335
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
337
const uint8_t *filter = subpel_filters[mx-1]; \
338
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
341
for (y = 0; y < h; y++) { \
342
for (x = 0; x < SIZE; x++) \
343
dst[x] = FILTER(src, filter, 1); \
348
#define VP8_EPEL_V(SIZE, FILTER, FILTERNAME) \
349
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
351
const uint8_t *filter = subpel_filters[my-1]; \
352
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
355
for (y = 0; y < h; y++) { \
356
for (x = 0; x < SIZE; x++) \
357
dst[x] = FILTER(src, filter, srcstride); \
362
#define VP8_EPEL_HV(SIZE, FILTERX, FILTERY, FILTERNAME) \
363
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
365
const uint8_t *filter = subpel_filters[mx-1]; \
366
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
368
uint8_t tmp_array[(2*SIZE+5)*SIZE]; \
369
uint8_t *tmp = tmp_array; \
370
src -= 2*srcstride; \
372
for (y = 0; y < h+5; y++) { \
373
for (x = 0; x < SIZE; x++) \
374
tmp[x] = FILTERX(src, filter, 1); \
379
tmp = tmp_array + 2*SIZE; \
380
filter = subpel_filters[my-1]; \
382
for (y = 0; y < h; y++) { \
383
for (x = 0; x < SIZE; x++) \
384
dst[x] = FILTERY(tmp, filter, SIZE); \
390
VP8_EPEL_H(16, FILTER_4TAP, h4)
391
VP8_EPEL_H(8, FILTER_4TAP, h4)
392
VP8_EPEL_H(4, FILTER_4TAP, h4)
393
VP8_EPEL_H(16, FILTER_6TAP, h6)
394
VP8_EPEL_H(8, FILTER_6TAP, h6)
395
VP8_EPEL_H(4, FILTER_6TAP, h6)
396
VP8_EPEL_V(16, FILTER_4TAP, v4)
397
VP8_EPEL_V(8, FILTER_4TAP, v4)
398
VP8_EPEL_V(4, FILTER_4TAP, v4)
399
VP8_EPEL_V(16, FILTER_6TAP, v6)
400
VP8_EPEL_V(8, FILTER_6TAP, v6)
401
VP8_EPEL_V(4, FILTER_6TAP, v6)
402
VP8_EPEL_HV(16, FILTER_4TAP, FILTER_4TAP, h4v4)
403
VP8_EPEL_HV(8, FILTER_4TAP, FILTER_4TAP, h4v4)
404
VP8_EPEL_HV(4, FILTER_4TAP, FILTER_4TAP, h4v4)
405
VP8_EPEL_HV(16, FILTER_4TAP, FILTER_6TAP, h4v6)
406
VP8_EPEL_HV(8, FILTER_4TAP, FILTER_6TAP, h4v6)
407
VP8_EPEL_HV(4, FILTER_4TAP, FILTER_6TAP, h4v6)
408
VP8_EPEL_HV(16, FILTER_6TAP, FILTER_4TAP, h6v4)
409
VP8_EPEL_HV(8, FILTER_6TAP, FILTER_4TAP, h6v4)
410
VP8_EPEL_HV(4, FILTER_6TAP, FILTER_4TAP, h6v4)
411
VP8_EPEL_HV(16, FILTER_6TAP, FILTER_6TAP, h6v6)
412
VP8_EPEL_HV(8, FILTER_6TAP, FILTER_6TAP, h6v6)
413
VP8_EPEL_HV(4, FILTER_6TAP, FILTER_6TAP, h6v6)
415
#define VP8_BILINEAR(SIZE) \
416
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
int a = 8-mx, b = mx; \
421
for (y = 0; y < h; y++) { \
422
for (x = 0; x < SIZE; x++) \
423
dst[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
428
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
int c = 8-my, d = my; \
433
for (y = 0; y < h; y++) { \
434
for (x = 0; x < SIZE; x++) \
435
dst[x] = (c*src[x] + d*src[x+stride] + 4) >> 3; \
441
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
int a = 8-mx, b = mx; \
444
int c = 8-my, d = my; \
446
uint8_t tmp_array[(2*SIZE+1)*SIZE]; \
447
uint8_t *tmp = tmp_array; \
449
for (y = 0; y < h+1; y++) { \
450
for (x = 0; x < SIZE; x++) \
451
tmp[x] = (a*src[x] + b*src[x+1] + 4) >> 3; \
458
for (y = 0; y < h; y++) { \
459
for (x = 0; x < SIZE; x++) \
460
dst[x] = (c*tmp[x] + d*tmp[x+SIZE] + 4) >> 3; \
470
#define VP8_MC_FUNC(IDX, SIZE) \
471
dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
472
dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \
473
dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \
474
dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \
475
dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
476
dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
477
dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \
478
dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
479
dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
481
#define VP8_BILINEAR_MC_FUNC(IDX, SIZE) \
482
dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \
483
dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
484
dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
485
dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
486
dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
487
dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
488
dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
489
dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
490
dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
492
av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
494
dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
495
dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
496
dsp->vp8_idct_add = vp8_idct_add_c;
497
dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
498
dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c;
499
dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
501
dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
502
dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
503
dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
504
dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
506
dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
507
dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
508
dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
509
dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
511
dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
512
dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
518
VP8_BILINEAR_MC_FUNC(0, 16);
519
VP8_BILINEAR_MC_FUNC(1, 8);
520
VP8_BILINEAR_MC_FUNC(2, 4);
523
ff_vp8dsp_init_x86(dsp);
525
ff_vp8dsp_init_altivec(dsp);