GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/cavs.c Lines: 470 487 96.5 %
Date: 2019-11-22 03:34:36 Branches: 210 231 90.9 %

Line Branch Exec Source
1
/*
2
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
3
 * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg 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
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
22
/**
23
 * @file
24
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
25
 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
26
 */
27
28
#include "avcodec.h"
29
#include "get_bits.h"
30
#include "golomb.h"
31
#include "h264chroma.h"
32
#include "idctdsp.h"
33
#include "internal.h"
34
#include "mathops.h"
35
#include "qpeldsp.h"
36
#include "cavs.h"
37
38
static const uint8_t alpha_tab[64] = {
39
     0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  2,  2,  2,  3,  3,
40
     4,  4,  5,  5,  6,  7,  8,  9, 10, 11, 12, 13, 15, 16, 18, 20,
41
    22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,
42
    46, 48, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64
43
};
44
45
static const uint8_t beta_tab[64] = {
46
     0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,
47
     2,  2,  3,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,  5,  6,  6,
48
     6,  7,  7,  7,  8,  8,  8,  9,  9, 10, 10, 11, 11, 12, 13, 14,
49
    15, 16, 17, 18, 19, 20, 21, 22, 23, 23, 24, 24, 25, 25, 26, 27
50
};
51
52
static const uint8_t tc_tab[64] = {
53
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
54
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
55
    2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,
56
    5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9, 9
57
};
58
59
/** mark block as unavailable, i.e. out of picture
60
 *  or not yet decoded */
61
static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL };
62
63
static const int8_t left_modifier_l[8] = {  0, -1,  6, -1, -1, 7, 6, 7 };
64
static const int8_t top_modifier_l[8]  = { -1,  1,  5, -1, -1, 5, 7, 7 };
65
static const int8_t left_modifier_c[7] = {  5, -1,  2, -1,  6, 5, 6 };
66
static const int8_t top_modifier_c[7]  = {  4,  1, -1, -1,  4, 6, 6 };
67
68
/*****************************************************************************
69
 *
70
 * in-loop deblocking filter
71
 *
72
 ****************************************************************************/
73
74
1793782
static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b)
75
{
76

1793782
    if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
77
15674
        return 2;
78
1778108
    if((abs(mvP->x - mvQ->x) >= 4) ||
79
1673007
       (abs(mvP->y - mvQ->y) >= 4) ||
80
1611712
       (mvP->ref != mvQ->ref))
81
303544
        return 1;
82
1474564
    if (b) {
83
1217805
        mvP += MV_BWD_OFFS;
84
1217805
        mvQ += MV_BWD_OFFS;
85
1217805
        if((abs(mvP->x - mvQ->x) >= 4) ||
86
1203525
           (abs(mvP->y - mvQ->y) >= 4) ||
87
1195077
           (mvP->ref != mvQ->ref))
88
67614
            return 1;
89
    }
90
1406950
    return 0;
91
}
92
93
#define SET_PARAMS                                                \
94
    alpha = alpha_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];  \
95
    beta  =  beta_tab[av_clip_uintp2(qp_avg + h->beta_offset,  6)];  \
96
    tc    =    tc_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];
97
98
/**
99
 * in-loop deblocking filter for a single macroblock
100
 *
101
 * boundary strength (bs) mapping:
102
 *
103
 * --4---5--
104
 * 0   2   |
105
 * | 6 | 7 |
106
 * 1   3   |
107
 * ---------
108
 */
109
297745
void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type)
110
{
111
    uint8_t bs[8];
112
    int qp_avg, alpha, beta, tc;
113
    int i;
114
115
    /* save un-deblocked lines */
116
297745
    h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15];
117
297745
    h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8];
118
297745
    h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8];
119
297745
    memcpy(&h->top_border_y[h->mbx * 16],     h->cy + 15 * h->l_stride, 16);
120
297745
    memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu +  7 * h->c_stride, 8);
121
297745
    memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv +  7 * h->c_stride, 8);
122
2679705
    for (i = 0; i < 8; i++) {
123
2381960
        h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride);
124
2381960
        h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride);
125
2381960
        h->left_border_u[i + 1]     = *(h->cu + 7  +  i          * h->c_stride);
126
2381960
        h->left_border_v[i + 1]     = *(h->cv + 7  +  i          * h->c_stride);
127
    }
128
297745
    if (!h->loop_filter_disable) {
129
        /* determine bs */
130
297745
        if (mb_type == I_8X8)
131
25344
            memset(bs, 2, 8);
132
        else {
133
272401
            memset(bs, 0, 8);
134
272401
            if (ff_cavs_partition_flags[mb_type] & SPLITV) {
135
175404
                bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
136
175404
                bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
137
            }
138
272401
            if (ff_cavs_partition_flags[mb_type] & SPLITH) {
139
176685
                bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
140
176685
                bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
141
            }
142
272401
            bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
143
272401
            bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
144
272401
            bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
145
272401
            bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
146
        }
147
297745
        if (AV_RN64(bs)) {
148
134358
            if (h->flags & A_AVAIL) {
149
127741
                qp_avg = (h->qp + h->left_qp + 1) >> 1;
150
127741
                SET_PARAMS;
151
127741
                h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]);
152
127741
                qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->left_qp] + 1) >> 1;
153
127741
                SET_PARAMS;
154
127741
                h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
155
127741
                h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
156
            }
157
134358
            qp_avg = h->qp;
158
134358
            SET_PARAMS;
159
134358
            h->cdsp.cavs_filter_lv(h->cy + 8,               h->l_stride, alpha, beta, tc, bs[2], bs[3]);
160
134358
            h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]);
161
162
134358
            if (h->flags & B_AVAIL) {
163
126078
                qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
164
126078
                SET_PARAMS;
165
126078
                h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]);
166
126078
                qp_avg = (ff_cavs_chroma_qp[h->qp] + ff_cavs_chroma_qp[h->top_qp[h->mbx]] + 1) >> 1;
167
126078
                SET_PARAMS;
168
126078
                h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
169
126078
                h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
170
            }
171
        }
172
    }
173
297745
    h->left_qp        = h->qp;
174
297745
    h->top_qp[h->mbx] = h->qp;
175
297745
}
176
177
#undef SET_PARAMS
178
179
/*****************************************************************************
180
 *
181
 * spatial intra prediction
182
 *
183
 ****************************************************************************/
184
185
101376
void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top,
186
                                  uint8_t **left, int block)
187
{
188
    int i;
189
190

101376
    switch (block) {
191
25344
    case 0:
192
25344
        *left               = h->left_border_y;
193
25344
        h->left_border_y[0] = h->left_border_y[1];
194
25344
        memset(&h->left_border_y[17], h->left_border_y[16], 9);
195
25344
        memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16);
196
25344
        top[17] = top[16];
197
25344
        top[0]  = top[1];
198

25344
        if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
199
24410
            h->left_border_y[0] = top[0] = h->topleft_border_y;
200
25344
        break;
201
25344
    case 1:
202
25344
        *left = h->intern_border_y;
203
228096
        for (i = 0; i < 8; i++)
204
202752
            h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride);
205
25344
        memset(&h->intern_border_y[9], h->intern_border_y[8], 9);
206
25344
        h->intern_border_y[0] = h->intern_border_y[1];
207
25344
        memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8);
208
25344
        if (h->flags & C_AVAIL)
209
24175
            memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8);
210
        else
211
1169
            memset(&top[9], top[8], 9);
212
25344
        top[17] = top[16];
213
25344
        top[0]  = top[1];
214
25344
        if (h->flags & B_AVAIL)
215
24808
            h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7];
216
25344
        break;
217
25344
    case 2:
218
25344
        *left = &h->left_border_y[8];
219
25344
        memcpy(&top[1], h->cy + 7 * h->l_stride, 16);
220
25344
        top[17] = top[16];
221
25344
        top[0]  = top[1];
222
25344
        if (h->flags & A_AVAIL)
223
24937
            top[0] = h->left_border_y[8];
224
25344
        break;
225
25344
    case 3:
226
25344
        *left = &h->intern_border_y[8];
227
228096
        for (i = 0; i < 8; i++)
228
202752
            h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride);
229
25344
        memset(&h->intern_border_y[17], h->intern_border_y[16], 9);
230
25344
        memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9);
231
25344
        memset(&top[9], top[8], 9);
232
25344
        break;
233
    }
234
101376
}
235
236
25344
void ff_cavs_load_intra_pred_chroma(AVSContext *h)
237
{
238
    /* extend borders by one pixel */
239
25344
    h->left_border_u[9]              = h->left_border_u[8];
240
25344
    h->left_border_v[9]              = h->left_border_v[8];
241
25344
    if(h->flags & C_AVAIL) {
242
24175
        h->top_border_u[h->mbx*10 + 9] = h->top_border_u[h->mbx*10 + 11];
243
24175
        h->top_border_v[h->mbx*10 + 9] = h->top_border_v[h->mbx*10 + 11];
244
    } else {
245
1169
        h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8];
246
1169
        h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8];
247
    }
248

25344
    if((h->flags & A_AVAIL) && (h->flags & B_AVAIL)) {
249
24410
        h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u;
250
24410
        h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v;
251
    } else {
252
934
        h->left_border_u[0]          = h->left_border_u[1];
253
934
        h->left_border_v[0]          = h->left_border_v[1];
254
934
        h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1];
255
934
        h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1];
256
    }
257
25344
}
258
259
45576
static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
260
{
261
    int y;
262
45576
    uint64_t a = AV_RN64(&top[1]);
263
410184
    for (y = 0; y < 8; y++)
264
364608
        *((uint64_t *)(d + y * stride)) = a;
265
45576
}
266
267
33618
static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
268
{
269
    int y;
270
    uint64_t a;
271
302562
    for (y = 0; y < 8; y++) {
272
268944
        a = left[y + 1] * 0x0101010101010101ULL;
273
268944
        *((uint64_t *)(d + y * stride)) = a;
274
    }
275
33618
}
276
277
27
static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
278
{
279
    int y;
280
27
    uint64_t a = 0x8080808080808080ULL;
281
243
    for (y = 0; y < 8; y++)
282
216
        *((uint64_t *)(d + y * stride)) = a;
283
27
}
284
285
2756
static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
286
{
287
    int x, y, ia;
288
2756
    int ih = 0;
289
2756
    int iv = 0;
290
2756
    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
291
292
13780
    for (x = 0; x < 4; x++) {
293
11024
        ih += (x + 1) *  (top[5 + x] -  top[3 - x]);
294
11024
        iv += (x + 1) * (left[5 + x] - left[3 - x]);
295
    }
296
2756
    ia = (top[8] + left[8]) << 4;
297
2756
    ih = (17 * ih + 16) >> 5;
298
2756
    iv = (17 * iv + 16) >> 5;
299
24804
    for (y = 0; y < 8; y++)
300
198432
        for (x = 0; x < 8; x++)
301
176384
            d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5];
302
2756
}
303
304
#define LOWPASS(ARRAY, INDEX)                                           \
305
    ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2)
306
307
61656
static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
308
{
309
    int x, y;
310
554904
    for (y = 0; y < 8; y++)
311
4439232
        for (x = 0; x < 8; x++)
312
3945984
            d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1;
313
61656
}
314
315
2887
static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
316
{
317
    int x, y;
318
25983
    for (y = 0; y < 8; y++)
319
207864
        for (x = 0; x < 8; x++)
320
184768
            d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1;
321
2887
}
322
323
1826
static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
324
{
325
    int x, y;
326
16434
    for (y = 0; y < 8; y++)
327
131472
        for (x = 0; x < 8; x++)
328
116864
            if (x == y)
329
14608
                d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2;
330
102256
            else if (x > y)
331
51128
                d[y * stride + x] = LOWPASS(top, x - y);
332
            else
333
51128
                d[y * stride + x] = LOWPASS(left, y - x);
334
1826
}
335
336
2117
static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
337
{
338
    int x, y;
339
19053
    for (y = 0; y < 8; y++)
340
152424
        for (x = 0; x < 8; x++)
341
135488
            d[y * stride + x] = LOWPASS(left, y + 1);
342
2117
}
343
344
1601
static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
345
{
346
    int x, y;
347
14409
    for (y = 0; y < 8; y++)
348
115272
        for (x = 0; x < 8; x++)
349
102464
            d[y * stride + x] = LOWPASS(top, x + 1);
350
1601
}
351
352
#undef LOWPASS
353
354
2829
static inline void modify_pred(const int8_t *mod_table, int *mode)
355
{
356
2829
    *mode = mod_table[*mode];
357
2829
    if (*mode < 0) {
358
        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
359
        *mode = 0;
360
    }
361
2829
}
362
363
25344
void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv)
364
{
365
    /* save pred modes before they get modified */
366
25344
    h->pred_mode_Y[3]             = h->pred_mode_Y[5];
367
25344
    h->pred_mode_Y[6]             = h->pred_mode_Y[8];
368
25344
    h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7];
369
25344
    h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8];
370
371
    /* modify pred modes according to availability of neighbour samples */
372
25344
    if (!(h->flags & A_AVAIL)) {
373
407
        modify_pred(left_modifier_l, &h->pred_mode_Y[4]);
374
407
        modify_pred(left_modifier_l, &h->pred_mode_Y[7]);
375
407
        modify_pred(left_modifier_c, pred_mode_uv);
376
    }
377
25344
    if (!(h->flags & B_AVAIL)) {
378
536
        modify_pred(top_modifier_l, &h->pred_mode_Y[4]);
379
536
        modify_pred(top_modifier_l, &h->pred_mode_Y[5]);
380
536
        modify_pred(top_modifier_c, pred_mode_uv);
381
    }
382
25344
}
383
384
/*****************************************************************************
385
 *
386
 * motion compensation
387
 *
388
 ****************************************************************************/
389
390
1448997
static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
391
                               int delta, int list, uint8_t *dest_y,
392
                               uint8_t *dest_cb, uint8_t *dest_cr,
393
                               int src_x_offset, int src_y_offset,
394
                               qpel_mc_func *qpix_op,
395
                               h264_chroma_mc_func chroma_op, cavs_vector *mv)
396
{
397
1448997
    const int mx         = mv->x + src_x_offset * 8;
398
1448997
    const int my         = mv->y + src_y_offset * 8;
399
1448997
    const int luma_xy    = (mx & 3) + ((my & 3) << 2);
400
1448997
    uint8_t *src_y       = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride;
401
1448997
    uint8_t *src_cb      = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride;
402
1448997
    uint8_t *src_cr      = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride;
403
1448997
    int extra_width      = 0;
404
1448997
    int extra_height     = extra_width;
405
1448997
    const int full_mx    = mx >> 2;
406
1448997
    const int full_my    = my >> 2;
407
1448997
    const int pic_width  = 16 * h->mb_width;
408
1448997
    const int pic_height = 16 * h->mb_height;
409
1448997
    int emu = 0;
410
411
1448997
    if (!pic->data[0])
412
        return;
413
1448997
    if (mx & 7)
414
120702
        extra_width  -= 3;
415
1448997
    if (my & 7)
416
125744
        extra_height -= 3;
417
418
1448997
    if (full_mx < 0 - extra_width ||
419
1448061
        full_my < 0 - extra_height ||
420
1446581
        full_mx + 16 /* FIXME */ > pic_width + extra_width ||
421
1431388
        full_my + 16 /* FIXME */ > pic_height + extra_height) {
422
37368
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
423
37368
                                 src_y - 2 - 2 * h->l_stride,
424
                                 h->l_stride, h->l_stride,
425
                                 16 + 5, 16 + 5 /* FIXME */,
426
                                 full_mx - 2, full_my - 2,
427
                                 pic_width, pic_height);
428
37368
        src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride;
429
37368
        emu   = 1;
430
    }
431
432
    // FIXME try variable height perhaps?
433
1448997
    qpix_op[luma_xy](dest_y, src_y, h->l_stride);
434
435
1448997
    if (emu) {
436
37368
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
437
                                 h->c_stride, h->c_stride,
438
                                 9, 9 /* FIXME */,
439
                                 mx >> 3, my >> 3,
440
                                 pic_width >> 1, pic_height >> 1);
441
37368
        src_cb = h->edge_emu_buffer;
442
    }
443
1448997
    chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7);
444
445
1448997
    if (emu) {
446
37368
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
447
                                 h->c_stride, h->c_stride,
448
                                 9, 9 /* FIXME */,
449
                                 mx >> 3, my >> 3,
450
                                 pic_width >> 1, pic_height >> 1);
451
37368
        src_cr = h->edge_emu_buffer;
452
    }
453
1448997
    chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7);
454
}
455
456
832240
static inline void mc_part_std(AVSContext *h, int chroma_height, int delta,
457
                               uint8_t *dest_y,
458
                               uint8_t *dest_cb,
459
                               uint8_t *dest_cr,
460
                               int x_offset, int y_offset,
461
                               qpel_mc_func *qpix_put,
462
                               h264_chroma_mc_func chroma_put,
463
                               qpel_mc_func *qpix_avg,
464
                               h264_chroma_mc_func chroma_avg,
465
                               cavs_vector *mv)
466
{
467
832240
    qpel_mc_func *qpix_op =  qpix_put;
468
832240
    h264_chroma_mc_func chroma_op = chroma_put;
469
470
832240
    dest_y   += x_offset * 2 + y_offset * h->l_stride * 2;
471
832240
    dest_cb  += x_offset     + y_offset * h->c_stride;
472
832240
    dest_cr  += x_offset     + y_offset * h->c_stride;
473
832240
    x_offset += 8 * h->mbx;
474
832240
    y_offset += 8 * h->mby;
475
476
832240
    if (mv->ref >= 0) {
477
788976
        AVFrame *ref = h->DPB[mv->ref].f;
478
788976
        mc_dir_part(h, ref, chroma_height, delta, 0,
479
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
480
                    qpix_op, chroma_op, mv);
481
482
788976
        qpix_op   = qpix_avg;
483
788976
        chroma_op = chroma_avg;
484
    }
485
486
832240
    if ((mv + MV_BWD_OFFS)->ref >= 0) {
487
660021
        AVFrame *ref = h->DPB[0].f;
488
660021
        mc_dir_part(h, ref, chroma_height, delta, 1,
489
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
490
                    qpix_op, chroma_op, mv + MV_BWD_OFFS);
491
    }
492
832240
}
493
494
272401
void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type)
495
{
496
272401
    if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16
497
85788
        mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0,
498
85788
                    h->cdsp.put_cavs_qpel_pixels_tab[0],
499
                    h->h264chroma.put_h264_chroma_pixels_tab[0],
500
85788
                    h->cdsp.avg_cavs_qpel_pixels_tab[0],
501
                    h->h264chroma.avg_h264_chroma_pixels_tab[0],
502
                    &h->mv[MV_FWD_X0]);
503
    } else {
504
186613
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0,
505
186613
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
506
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
507
186613
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
508
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
509
                    &h->mv[MV_FWD_X0]);
510
186613
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0,
511
186613
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
512
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
513
186613
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
514
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
515
                    &h->mv[MV_FWD_X1]);
516
186613
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4,
517
186613
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
518
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
519
186613
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
520
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
521
                    &h->mv[MV_FWD_X2]);
522
186613
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4,
523
186613
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
524
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
525
186613
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
526
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
527
                    &h->mv[MV_FWD_X3]);
528
    }
529
272401
}
530
531
/*****************************************************************************
532
 *
533
 * motion vector prediction
534
 *
535
 ****************************************************************************/
536
537
510849
static inline void scale_mv(AVSContext *h, int *d_x, int *d_y,
538
                            cavs_vector *src, int distp)
539
{
540
510849
    int64_t den = h->scale_den[FFMAX(src->ref, 0)];
541
510849
    *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9;
542
510849
    *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9;
543
510849
}
544
545
170283
static inline void mv_pred_median(AVSContext *h,
546
                                  cavs_vector *mvP,
547
                                  cavs_vector *mvA,
548
                                  cavs_vector *mvB,
549
                                  cavs_vector *mvC)
550
{
551
    int ax, ay, bx, by, cx, cy;
552
    int len_ab, len_bc, len_ca, len_mid;
553
554
    /* scale candidates according to their temporal span */
555
170283
    scale_mv(h, &ax, &ay, mvA, mvP->dist);
556
170283
    scale_mv(h, &bx, &by, mvB, mvP->dist);
557
170283
    scale_mv(h, &cx, &cy, mvC, mvP->dist);
558
    /* find the geometrical median of the three candidates */
559
170283
    len_ab  = abs(ax - bx) + abs(ay - by);
560
170283
    len_bc  = abs(bx - cx) + abs(by - cy);
561
170283
    len_ca  = abs(cx - ax) + abs(cy - ay);
562
170283
    len_mid = mid_pred(len_ab, len_bc, len_ca);
563
170283
    if (len_mid == len_ab) {
564
137907
        mvP->x = cx;
565
137907
        mvP->y = cy;
566
32376
    } else if (len_mid == len_bc) {
567
24630
        mvP->x = ax;
568
24630
        mvP->y = ay;
569
    } else {
570
7746
        mvP->x = bx;
571
7746
        mvP->y = by;
572
    }
573
170283
}
574
575
238918
void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,
576
                enum cavs_mv_pred mode, enum cavs_block size, int ref)
577
{
578
238918
    cavs_vector *mvP = &h->mv[nP];
579
238918
    cavs_vector *mvA = &h->mv[nP-1];
580
238918
    cavs_vector *mvB = &h->mv[nP-4];
581
238918
    cavs_vector *mvC = &h->mv[nC];
582
238918
    const cavs_vector *mvP2 = NULL;
583
584
238918
    mvP->ref  = ref;
585
238918
    mvP->dist = h->dist[mvP->ref];
586

238918
    if (mvC->ref == NOT_AVAIL || (nP == MV_FWD_X3) || (nP == MV_BWD_X3 ))
587
28240
        mvC = &h->mv[nP - 5];  // set to top-left (mvD)
588
238918
    if (mode == MV_PRED_PSKIP &&
589
20801
        (mvA->ref == NOT_AVAIL ||
590
20081
         mvB->ref == NOT_AVAIL ||
591
19308
         (mvA->x | mvA->y | mvA->ref) == 0 ||
592
850
         (mvB->x | mvB->y | mvB->ref) == 0)) {
593
20434
        mvP2 = &un_mv;
594
    /* if there is only one suitable candidate, take it */
595

218484
    } else if (mvA->ref >= 0 && mvB->ref < 0  && mvC->ref < 0) {
596
11224
        mvP2 = mvA;
597

207260
    } else if (mvA->ref < 0  && mvB->ref >= 0 && mvC->ref < 0) {
598
4577
        mvP2 = mvB;
599

202683
    } else if (mvA->ref < 0  && mvB->ref < 0  && mvC->ref >= 0) {
600
3525
        mvP2 = mvC;
601

199158
    } else if (mode == MV_PRED_LEFT     && mvA->ref == ref) {
602
14269
        mvP2 = mvA;
603

184889
    } else if (mode == MV_PRED_TOP      && mvB->ref == ref) {
604
7898
        mvP2 = mvB;
605

176991
    } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) {
606
6708
        mvP2 = mvC;
607
    }
608
238918
    if (mvP2) {
609
68635
        mvP->x = mvP2->x;
610
68635
        mvP->y = mvP2->y;
611
    } else
612
170283
        mv_pred_median(h, mvP, mvA, mvB, mvC);
613
614
238918
    if (mode < MV_PRED_PSKIP) {
615
189273
        int mx = get_se_golomb(&h->gb) + (unsigned)mvP->x;
616
189273
        int my = get_se_golomb(&h->gb) + (unsigned)mvP->y;
617
618

189273
        if (mx != (int16_t)mx || my != (int16_t)my) {
619
            av_log(h->avctx, AV_LOG_ERROR, "MV %d %d out of supported range\n", mx, my);
620
        } else {
621
189273
            mvP->x = mx;
622
189273
            mvP->y = my;
623
        }
624
    }
625
238918
    set_mvs(mvP, size);
626
238918
}
627
628
/*****************************************************************************
629
 *
630
 * macroblock level
631
 *
632
 ****************************************************************************/
633
634
/**
635
 * initialise predictors for motion vectors and intra prediction
636
 */
637
297745
void ff_cavs_init_mb(AVSContext *h)
638
{
639
    int i;
640
641
    /* copy predictors from top line (MB B and C) into cache */
642
1190980
    for (i = 0; i < 3; i++) {
643
893235
        h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i];
644
893235
        h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i];
645
    }
646
297745
    h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0];
647
297745
    h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1];
648
    /* clear top predictors if MB B is not available */
649
297745
    if (!(h->flags & B_AVAIL)) {
650
8280
        h->mv[MV_FWD_B2]  = un_mv;
651
8280
        h->mv[MV_FWD_B3]  = un_mv;
652
8280
        h->mv[MV_BWD_B2]  = un_mv;
653
8280
        h->mv[MV_BWD_B3]  = un_mv;
654
8280
        h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
655
8280
        h->flags         &= ~(C_AVAIL | D_AVAIL);
656
289465
    } else if (h->mbx) {
657
283032
        h->flags |= D_AVAIL;
658
    }
659
297745
    if (h->mbx == h->mb_width - 1) // MB C not available
660
6616
        h->flags &= ~C_AVAIL;
661
    /* clear top-right predictors if MB C is not available */
662
297745
    if (!(h->flags & C_AVAIL)) {
663
14712
        h->mv[MV_FWD_C2] = un_mv;
664
14712
        h->mv[MV_BWD_C2] = un_mv;
665
    }
666
    /* clear top-left predictors if MB D is not available */
667
297745
    if (!(h->flags & D_AVAIL)) {
668
14713
        h->mv[MV_FWD_D3] = un_mv;
669
14713
        h->mv[MV_BWD_D3] = un_mv;
670
    }
671
297745
}
672
673
/**
674
 * save predictors for later macroblocks and increase
675
 * macroblock address
676
 * @return 0 if end of frame is reached, 1 otherwise
677
 */
678
297745
int ff_cavs_next_mb(AVSContext *h)
679
{
680
    int i;
681
682
297745
    h->flags |= A_AVAIL;
683
297745
    h->cy    += 16;
684
297745
    h->cu    += 8;
685
297745
    h->cv    += 8;
686
    /* copy mvs as predictors to the left */
687
2084215
    for (i = 0; i <= 20; i += 4)
688
1786470
        h->mv[i] = h->mv[i + 2];
689
    /* copy bottom mvs from cache to top line */
690
297745
    h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2];
691
297745
    h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3];
692
297745
    h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2];
693
297745
    h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3];
694
    /* next MB address */
695
297745
    h->mbidx++;
696
297745
    h->mbx++;
697
297745
    if (h->mbx == h->mb_width) { // New mb line
698
6616
        h->flags = B_AVAIL | C_AVAIL;
699
        /* clear left pred_modes */
700
6616
        h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
701
        /* clear left mv predictors */
702
46312
        for (i = 0; i <= 20; i += 4)
703
39696
            h->mv[i] = un_mv;
704
6616
        h->mbx = 0;
705
6616
        h->mby++;
706
        /* re-calculate sample pointers */
707
6616
        h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride;
708
6616
        h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride;
709
6616
        h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride;
710
6616
        if (h->mby == h->mb_height) { // Frame end
711
183
            return 0;
712
        }
713
    }
714
297562
    return 1;
715
}
716
717
/*****************************************************************************
718
 *
719
 * frame level
720
 *
721
 ****************************************************************************/
722
723
184
int ff_cavs_init_pic(AVSContext *h)
724
{
725
    int i;
726
727
    /* clear some predictors */
728
1288
    for (i = 0; i <= 20; i += 4)
729
1104
        h->mv[i] = un_mv;
730
184
    h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
731
184
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
732
184
    h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
733
184
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
734
184
    h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
735
184
    h->cy             = h->cur.f->data[0];
736
184
    h->cu             = h->cur.f->data[1];
737
184
    h->cv             = h->cur.f->data[2];
738
184
    h->l_stride       = h->cur.f->linesize[0];
739
184
    h->c_stride       = h->cur.f->linesize[1];
740
184
    h->luma_scan[2]   = 8 * h->l_stride;
741
184
    h->luma_scan[3]   = 8 * h->l_stride + 8;
742
184
    h->mbx            = h->mby = h->mbidx = 0;
743
184
    h->flags          = 0;
744
745
184
    return 0;
746
}
747
748
/*****************************************************************************
749
 *
750
 * headers and interface
751
 *
752
 ****************************************************************************/
753
754
/**
755
 * some predictions require data from the top-neighbouring macroblock.
756
 * this data has to be stored for one complete row of macroblocks
757
 * and this storage space is allocated here
758
 */
759
4
int ff_cavs_init_top_lines(AVSContext *h)
760
{
761
    /* alloc top line of predictors */
762
4
    h->top_qp       = av_mallocz(h->mb_width);
763
4
    h->top_mv[0]    = av_mallocz_array(h->mb_width * 2 + 1,  sizeof(cavs_vector));
764
4
    h->top_mv[1]    = av_mallocz_array(h->mb_width * 2 + 1,  sizeof(cavs_vector));
765
4
    h->top_pred_Y   = av_mallocz_array(h->mb_width * 2,  sizeof(*h->top_pred_Y));
766
4
    h->top_border_y = av_mallocz_array(h->mb_width + 1,  16);
767
4
    h->top_border_u = av_mallocz_array(h->mb_width,  10);
768
4
    h->top_border_v = av_mallocz_array(h->mb_width,  10);
769
770
    /* alloc space for co-located MVs and types */
771
4
    h->col_mv        = av_mallocz_array(h->mb_width * h->mb_height,
772
                                        4 * sizeof(cavs_vector));
773
4
    h->col_type_base = av_mallocz(h->mb_width * h->mb_height);
774
4
    h->block         = av_mallocz(64 * sizeof(int16_t));
775
776


4
    if (!h->top_qp || !h->top_mv[0] || !h->top_mv[1] || !h->top_pred_Y ||
777

4
        !h->top_border_y || !h->top_border_u || !h->top_border_v ||
778

4
        !h->col_mv || !h->col_type_base || !h->block) {
779
        av_freep(&h->top_qp);
780
        av_freep(&h->top_mv[0]);
781
        av_freep(&h->top_mv[1]);
782
        av_freep(&h->top_pred_Y);
783
        av_freep(&h->top_border_y);
784
        av_freep(&h->top_border_u);
785
        av_freep(&h->top_border_v);
786
        av_freep(&h->col_mv);
787
        av_freep(&h->col_type_base);
788
        av_freep(&h->block);
789
        return AVERROR(ENOMEM);
790
    }
791
4
    return 0;
792
}
793
794
4
av_cold int ff_cavs_init(AVCodecContext *avctx)
795
{
796
4
    AVSContext *h = avctx->priv_data;
797
798
4
    ff_blockdsp_init(&h->bdsp, avctx);
799
4
    ff_h264chroma_init(&h->h264chroma, 8);
800
4
    ff_idctdsp_init(&h->idsp, avctx);
801
4
    ff_videodsp_init(&h->vdsp, 8);
802
4
    ff_cavsdsp_init(&h->cdsp, avctx);
803
4
    ff_init_scantable_permutation(h->idsp.idct_permutation,
804
4
                                  h->cdsp.idct_perm);
805
4
    ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct);
806
807
4
    h->avctx       = avctx;
808
4
    avctx->pix_fmt = AV_PIX_FMT_YUV420P;
809
810
4
    h->cur.f    = av_frame_alloc();
811
4
    h->DPB[0].f = av_frame_alloc();
812
4
    h->DPB[1].f = av_frame_alloc();
813

4
    if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f) {
814
        ff_cavs_end(avctx);
815
        return AVERROR(ENOMEM);
816
    }
817
818
4
    h->luma_scan[0]                     = 0;
819
4
    h->luma_scan[1]                     = 8;
820
4
    h->intra_pred_l[INTRA_L_VERT]       = intra_pred_vert;
821
4
    h->intra_pred_l[INTRA_L_HORIZ]      = intra_pred_horiz;
822
4
    h->intra_pred_l[INTRA_L_LP]         = intra_pred_lp;
823
4
    h->intra_pred_l[INTRA_L_DOWN_LEFT]  = intra_pred_down_left;
824
4
    h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
825
4
    h->intra_pred_l[INTRA_L_LP_LEFT]    = intra_pred_lp_left;
826
4
    h->intra_pred_l[INTRA_L_LP_TOP]     = intra_pred_lp_top;
827
4
    h->intra_pred_l[INTRA_L_DC_128]     = intra_pred_dc_128;
828
4
    h->intra_pred_c[INTRA_C_LP]         = intra_pred_lp;
829
4
    h->intra_pred_c[INTRA_C_HORIZ]      = intra_pred_horiz;
830
4
    h->intra_pred_c[INTRA_C_VERT]       = intra_pred_vert;
831
4
    h->intra_pred_c[INTRA_C_PLANE]      = intra_pred_plane;
832
4
    h->intra_pred_c[INTRA_C_LP_LEFT]    = intra_pred_lp_left;
833
4
    h->intra_pred_c[INTRA_C_LP_TOP]     = intra_pred_lp_top;
834
4
    h->intra_pred_c[INTRA_C_DC_128]     = intra_pred_dc_128;
835
4
    h->mv[7]                            = un_mv;
836
4
    h->mv[19]                           = un_mv;
837
4
    return 0;
838
}
839
840
4
av_cold int ff_cavs_end(AVCodecContext *avctx)
841
{
842
4
    AVSContext *h = avctx->priv_data;
843
844
4
    av_frame_free(&h->cur.f);
845
4
    av_frame_free(&h->DPB[0].f);
846
4
    av_frame_free(&h->DPB[1].f);
847
848
4
    av_freep(&h->top_qp);
849
4
    av_freep(&h->top_mv[0]);
850
4
    av_freep(&h->top_mv[1]);
851
4
    av_freep(&h->top_pred_Y);
852
4
    av_freep(&h->top_border_y);
853
4
    av_freep(&h->top_border_u);
854
4
    av_freep(&h->top_border_v);
855
4
    av_freep(&h->col_mv);
856
4
    av_freep(&h->col_type_base);
857
4
    av_freep(&h->block);
858
4
    av_freep(&h->edge_emu_buffer);
859
4
    return 0;
860
}