cavs.c 29.4 KB
Newer Older
1 2 3 4
/*
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
 * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * 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 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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
 */

Stefan Gehrer's avatar
Stefan Gehrer committed
22
/**
23
 * @file
Stefan Gehrer's avatar
Stefan Gehrer committed
24 25 26 27
 * Chinese AVS video (AVS1-P2, JiZhun profile) decoder
 * @author Stefan Gehrer <stefan.gehrer@gmx.de>
 */

28
#include "avcodec.h"
29 30
#include "bitstream.h"
#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 40 41 42
     0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  2,  2,  2,  3,  3,
     4,  4,  5,  5,  6,  7,  8,  9, 10, 11, 12, 13, 15, 16, 18, 20,
    22, 24, 26, 28, 30, 33, 33, 35, 35, 36, 37, 37, 39, 39, 42, 44,
    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 47 48 49
     0,  0,  0,  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  2,  2,  2,
     2,  2,  3,  3,  3,  3,  4,  4,  4,  4,  5,  5,  5,  5,  6,  6,
     6,  7,  7,  7,  8,  8,  8,  9,  9, 10, 10, 11, 11, 12, 13, 14,
    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 54 55 56
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
    2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4,
    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 62 63 64 65 66 67
static const cavs_vector un_mv = { 0, 0, 1, NOT_AVAIL };

static const int8_t left_modifier_l[8] = {  0, -1,  6, -1, -1, 7, 6, 7 };
static const int8_t top_modifier_l[8]  = { -1,  1,  5, -1, -1, 5, 7, 7 };
static const int8_t left_modifier_c[7] = {  5, -1,  2, -1,  6, 5, 6 };
static const int8_t top_modifier_c[7]  = {  4,  1, -1, -1,  4, 6, 6 };

68 69 70 71 72 73
/*****************************************************************************
 *
 * in-loop deblocking filter
 *
 ****************************************************************************/

74 75 76
static inline int get_bs(cavs_vector *mvP, cavs_vector *mvQ, int b)
{
    if ((mvP->ref == REF_INTRA) || (mvQ->ref == REF_INTRA))
77
        return 2;
78
    if ((abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4))
79
        return 1;
80
    if (b) {
Michael Niedermayer's avatar
Michael Niedermayer committed
81 82
        mvP += MV_BWD_OFFS;
        mvQ += MV_BWD_OFFS;
83
        if ((abs(mvP->x - mvQ->x) >= 4) || (abs(mvP->y - mvQ->y) >= 4))
Michael Niedermayer's avatar
Michael Niedermayer committed
84
            return 1;
85 86
    } else {
        if (mvP->ref != mvQ->ref)
87 88 89 90 91
            return 1;
    }
    return 0;
}

92
#define SET_PARAMS                                                \
93 94 95
    alpha = alpha_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];  \
    beta  =  beta_tab[av_clip_uintp2(qp_avg + h->beta_offset,  6)];  \
    tc    =    tc_tab[av_clip_uintp2(qp_avg + h->alpha_offset, 6)];
Stefan Gehrer's avatar
Stefan Gehrer committed
96 97 98 99 100

/**
 * in-loop deblocking filter for a single macroblock
 *
 * boundary strength (bs) mapping:
101 102 103 104 105 106 107
 *
 * --4---5--
 * 0   2   |
 * | 6 | 7 |
 * 1   3   |
 * ---------
 */
108 109
void ff_cavs_filter(AVSContext *h, enum cavs_mb mb_type)
{
110
    uint8_t bs[8];
111 112 113 114
    int qp_avg, alpha, beta, tc;
    int i;

    /* save un-deblocked lines */
115 116 117 118 119 120 121 122 123 124 125
    h->topleft_border_y = h->top_border_y[h->mbx * 16 + 15];
    h->topleft_border_u = h->top_border_u[h->mbx * 10 + 8];
    h->topleft_border_v = h->top_border_v[h->mbx * 10 + 8];
    memcpy(&h->top_border_y[h->mbx * 16],     h->cy + 15 * h->l_stride, 16);
    memcpy(&h->top_border_u[h->mbx * 10 + 1], h->cu +  7 * h->c_stride, 8);
    memcpy(&h->top_border_v[h->mbx * 10 + 1], h->cv +  7 * h->c_stride, 8);
    for (i = 0; i < 8; i++) {
        h->left_border_y[i * 2 + 1] = *(h->cy + 15 + (i * 2 + 0) * h->l_stride);
        h->left_border_y[i * 2 + 2] = *(h->cy + 15 + (i * 2 + 1) * h->l_stride);
        h->left_border_u[i + 1]     = *(h->cu + 7  +  i          * h->c_stride);
        h->left_border_v[i + 1]     = *(h->cv + 7  +  i          * h->c_stride);
126
    }
127
    if (!h->loop_filter_disable) {
128
        /* determine bs */
129 130
        if (mb_type == I_8X8)
            memset(bs, 2, 8);
131
        else {
132 133
            memset(bs, 0, 8);
            if (ff_cavs_partition_flags[mb_type] & SPLITV) {
Michael Niedermayer's avatar
Michael Niedermayer committed
134 135
                bs[2] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X1], mb_type > P_8X8);
                bs[3] = get_bs(&h->mv[MV_FWD_X2], &h->mv[MV_FWD_X3], mb_type > P_8X8);
136
            }
137
            if (ff_cavs_partition_flags[mb_type] & SPLITH) {
Michael Niedermayer's avatar
Michael Niedermayer committed
138 139 140 141 142 143 144
                bs[6] = get_bs(&h->mv[MV_FWD_X0], &h->mv[MV_FWD_X2], mb_type > P_8X8);
                bs[7] = get_bs(&h->mv[MV_FWD_X1], &h->mv[MV_FWD_X3], mb_type > P_8X8);
            }
            bs[0] = get_bs(&h->mv[MV_FWD_A1], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[1] = get_bs(&h->mv[MV_FWD_A3], &h->mv[MV_FWD_X2], mb_type > P_8X8);
            bs[4] = get_bs(&h->mv[MV_FWD_B2], &h->mv[MV_FWD_X0], mb_type > P_8X8);
            bs[5] = get_bs(&h->mv[MV_FWD_B3], &h->mv[MV_FWD_X1], mb_type > P_8X8);
145
        }
146 147
        if (AV_RN64(bs)) {
            if (h->flags & A_AVAIL) {
148 149
                qp_avg = (h->qp + h->left_qp + 1) >> 1;
                SET_PARAMS;
150 151 152
                h->cdsp.cavs_filter_lv(h->cy, h->l_stride, alpha, beta, tc, bs[0], bs[1]);
                h->cdsp.cavs_filter_cv(h->cu, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
                h->cdsp.cavs_filter_cv(h->cv, h->c_stride, alpha, beta, tc, bs[0], bs[1]);
153 154 155
            }
            qp_avg = h->qp;
            SET_PARAMS;
156 157
            h->cdsp.cavs_filter_lv(h->cy + 8,               h->l_stride, alpha, beta, tc, bs[2], bs[3]);
            h->cdsp.cavs_filter_lh(h->cy + 8 * h->l_stride, h->l_stride, alpha, beta, tc, bs[6], bs[7]);
158

159
            if (h->flags & B_AVAIL) {
160 161
                qp_avg = (h->qp + h->top_qp[h->mbx] + 1) >> 1;
                SET_PARAMS;
162 163 164
                h->cdsp.cavs_filter_lh(h->cy, h->l_stride, alpha, beta, tc, bs[4], bs[5]);
                h->cdsp.cavs_filter_ch(h->cu, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
                h->cdsp.cavs_filter_ch(h->cv, h->c_stride, alpha, beta, tc, bs[4], bs[5]);
165 166 167
            }
        }
    }
168
    h->left_qp        = h->qp;
169 170 171 172 173 174 175 176 177 178 179
    h->top_qp[h->mbx] = h->qp;
}

#undef SET_PARAMS

/*****************************************************************************
 *
 * spatial intra prediction
 *
 ****************************************************************************/

180
void ff_cavs_load_intra_pred_luma(AVSContext *h, uint8_t *top,
181 182
                                  uint8_t **left, int block)
{
183 184
    int i;

185
    switch (block) {
186
    case 0:
187
        *left               = h->left_border_y;
188
        h->left_border_y[0] = h->left_border_y[1];
189 190
        memset(&h->left_border_y[17], h->left_border_y[16], 9);
        memcpy(&top[1], &h->top_border_y[h->mbx * 16], 16);
191
        top[17] = top[16];
192 193
        top[0]  = top[1];
        if ((h->flags & A_AVAIL) && (h->flags & B_AVAIL))
194 195 196 197
            h->left_border_y[0] = top[0] = h->topleft_border_y;
        break;
    case 1:
        *left = h->intern_border_y;
198 199 200
        for (i = 0; i < 8; i++)
            h->intern_border_y[i + 1] = *(h->cy + 7 + i * h->l_stride);
        memset(&h->intern_border_y[9], h->intern_border_y[8], 9);
201
        h->intern_border_y[0] = h->intern_border_y[1];
202 203 204
        memcpy(&top[1], &h->top_border_y[h->mbx * 16 + 8], 8);
        if (h->flags & C_AVAIL)
            memcpy(&top[9], &h->top_border_y[(h->mbx + 1) * 16], 8);
205
        else
206
            memset(&top[9], top[8], 9);
207
        top[17] = top[16];
208 209 210
        top[0]  = top[1];
        if (h->flags & B_AVAIL)
            h->intern_border_y[0] = top[0] = h->top_border_y[h->mbx * 16 + 7];
211 212 213
        break;
    case 2:
        *left = &h->left_border_y[8];
214
        memcpy(&top[1], h->cy + 7 * h->l_stride, 16);
215
        top[17] = top[16];
216 217
        top[0]  = top[1];
        if (h->flags & A_AVAIL)
218 219 220 221
            top[0] = h->left_border_y[8];
        break;
    case 3:
        *left = &h->intern_border_y[8];
222 223 224 225 226
        for (i = 0; i < 8; i++)
            h->intern_border_y[i + 9] = *(h->cy + 7 + (i + 8) * h->l_stride);
        memset(&h->intern_border_y[17], h->intern_border_y[16], 9);
        memcpy(&top[0], h->cy + 7 + 7 * h->l_stride, 9);
        memset(&top[9], top[8], 9);
227 228 229 230
        break;
    }
}

231 232
void ff_cavs_load_intra_pred_chroma(AVSContext *h)
{
233
    /* extend borders by one pixel */
234 235
    h->left_border_u[9]              = h->left_border_u[8];
    h->left_border_v[9]              = h->left_border_v[8];
236 237 238 239 240
    h->top_border_u[h->mbx * 10 + 9] = h->top_border_u[h->mbx * 10 + 8];
    h->top_border_v[h->mbx * 10 + 9] = h->top_border_v[h->mbx * 10 + 8];
    if (h->mbx && h->mby) {
        h->top_border_u[h->mbx * 10] = h->left_border_u[0] = h->topleft_border_u;
        h->top_border_v[h->mbx * 10] = h->left_border_v[0] = h->topleft_border_v;
241
    } else {
242 243
        h->left_border_u[0]          = h->left_border_u[1];
        h->left_border_v[0]          = h->left_border_v[1];
244 245
        h->top_border_u[h->mbx * 10] = h->top_border_u[h->mbx * 10 + 1];
        h->top_border_v[h->mbx * 10] = h->top_border_v[h->mbx * 10 + 1];
246 247 248
    }
}

249
static void intra_pred_vert(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
250
{
251
    int y;
252
    uint64_t a = AV_RN64(&top[1]);
253
    for (y = 0; y < 8; y++)
254
        *((uint64_t *)(d + y * stride)) = a;
255 256
}

257
static void intra_pred_horiz(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
258
{
259 260
    int y;
    uint64_t a;
261 262 263
    for (y = 0; y < 8; y++) {
        a = left[y + 1] * 0x0101010101010101ULL;
        *((uint64_t *)(d + y * stride)) = a;
264 265 266
    }
}

267
static void intra_pred_dc_128(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
268
{
269 270
    int y;
    uint64_t a = 0x8080808080808080ULL;
271 272
    for (y = 0; y < 8; y++)
        *((uint64_t *)(d + y * stride)) = a;
273 274
}

275
static void intra_pred_plane(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
276 277
{
    int x, y, ia;
278 279
    int ih = 0;
    int iv = 0;
280
    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
281

282
    for (x = 0; x < 4; x++) {
283
        ih += (x + 1) *  (top[5 + x] -  top[3 - x]);
284
        iv += (x + 1) * (left[5 + x] - left[3 - x]);
285
    }
286 287 288 289 290 291
    ia = (top[8] + left[8]) << 4;
    ih = (17 * ih + 16) >> 5;
    iv = (17 * iv + 16) >> 5;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            d[y * stride + x] = cm[(ia + (x - 3) * ih + (y - 3) * iv + 16) >> 5];
292 293
}

294
#define LOWPASS(ARRAY, INDEX)                                           \
295
    ((ARRAY[(INDEX) - 1] + 2 * ARRAY[(INDEX)] + ARRAY[(INDEX) + 1] + 2) >> 2)
296

297
static void intra_pred_lp(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
298 299 300 301 302
{
    int x, y;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            d[y * stride + x] = (LOWPASS(top, x + 1) + LOWPASS(left, y + 1)) >> 1;
303 304
}

305
static void intra_pred_down_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
306 307 308 309 310
{
    int x, y;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            d[y * stride + x] = (LOWPASS(top, x + y + 2) + LOWPASS(left, x + y + 2)) >> 1;
311 312
}

313
static void intra_pred_down_right(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
314 315 316 317 318 319 320 321
{
    int x, y;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            if (x == y)
                d[y * stride + x] = (left[1] + 2 * top[0] + top[1] + 2) >> 2;
            else if (x > y)
                d[y * stride + x] = LOWPASS(top, x - y);
322
            else
323
                d[y * stride + x] = LOWPASS(left, y - x);
324 325
}

326
static void intra_pred_lp_left(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
327 328 329 330 331
{
    int x, y;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            d[y * stride + x] = LOWPASS(left, y + 1);
332 333
}

334
static void intra_pred_lp_top(uint8_t *d, uint8_t *top, uint8_t *left, ptrdiff_t stride)
335 336 337 338 339
{
    int x, y;
    for (y = 0; y < 8; y++)
        for (x = 0; x < 8; x++)
            d[y * stride + x] = LOWPASS(top, x + 1);
340 341 342 343
}

#undef LOWPASS

344 345 346
static inline void modify_pred(const int8_t *mod_table, int *mode)
{
    *mode = mod_table[*mode];
347
    if (*mode < 0) {
348 349 350 351 352
        av_log(NULL, AV_LOG_ERROR, "Illegal intra prediction mode\n");
        *mode = 0;
    }
}

353 354
void ff_cavs_modify_mb_i(AVSContext *h, int *pred_mode_uv)
{
355
    /* save pred modes before they get modified */
356 357
    h->pred_mode_Y[3]             = h->pred_mode_Y[5];
    h->pred_mode_Y[6]             = h->pred_mode_Y[8];
358 359
    h->top_pred_Y[h->mbx * 2 + 0] = h->pred_mode_Y[7];
    h->top_pred_Y[h->mbx * 2 + 1] = h->pred_mode_Y[8];
360 361

    /* modify pred modes according to availability of neighbour samples */
362
    if (!(h->flags & A_AVAIL)) {
363 364 365
        modify_pred(left_modifier_l, &h->pred_mode_Y[4]);
        modify_pred(left_modifier_l, &h->pred_mode_Y[7]);
        modify_pred(left_modifier_c, pred_mode_uv);
366
    }
367
    if (!(h->flags & B_AVAIL)) {
368 369 370
        modify_pred(top_modifier_l, &h->pred_mode_Y[4]);
        modify_pred(top_modifier_l, &h->pred_mode_Y[5]);
        modify_pred(top_modifier_c, pred_mode_uv);
371 372 373
    }
}

374 375 376 377 378 379
/*****************************************************************************
 *
 * motion compensation
 *
 ****************************************************************************/

380 381 382 383 384 385
static inline void mc_dir_part(AVSContext *h, AVFrame *pic, int chroma_height,
                               int delta, int list, uint8_t *dest_y,
                               uint8_t *dest_cb, uint8_t *dest_cr,
                               int src_x_offset, int src_y_offset,
                               qpel_mc_func *qpix_op,
                               h264_chroma_mc_func chroma_op, cavs_vector *mv)
Aneesh Dogra's avatar
Aneesh Dogra committed
386
{
387 388
    const int mx         = mv->x + src_x_offset * 8;
    const int my         = mv->y + src_y_offset * 8;
389 390 391 392 393 394 395 396 397 398 399
    const int luma_xy    = (mx & 3) + ((my & 3) << 2);
    uint8_t *src_y       = pic->data[0] + (mx >> 2) + (my >> 2) * h->l_stride;
    uint8_t *src_cb      = pic->data[1] + (mx >> 3) + (my >> 3) * h->c_stride;
    uint8_t *src_cr      = pic->data[2] + (mx >> 3) + (my >> 3) * h->c_stride;
    int extra_width      = 0;
    int extra_height     = extra_width;
    const int full_mx    = mx >> 2;
    const int full_my    = my >> 2;
    const int pic_width  = 16 * h->mb_width;
    const int pic_height = 16 * h->mb_height;
    int emu = 0;
400

401
    if (!pic->data[0])
402
        return;
403 404 405 406 407 408 409 410 411 412
    if (mx & 7)
        extra_width  -= 3;
    if (my & 7)
        extra_height -= 3;

    if (full_mx < 0 - extra_width ||
        full_my < 0 - extra_height ||
        full_mx + 16 /* FIXME */ > pic_width + extra_width ||
        full_my + 16 /* FIXME */ > pic_height + extra_height) {
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
413 414
                                 src_y - 2 - 2 * h->l_stride,
                                 h->l_stride, h->l_stride,
415 416 417 418
                                 16 + 5, 16 + 5 /* FIXME */,
                                 full_mx - 2, full_my - 2,
                                 pic_width, pic_height);
        src_y = h->edge_emu_buffer + 2 + 2 * h->l_stride;
419
        emu   = 1;
420 421
    }

422 423
    // FIXME try variable height perhaps?
    qpix_op[luma_xy](dest_y, src_y, h->l_stride);
424

425 426
    if (emu) {
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
427
                                 h->c_stride, h->c_stride,
428 429 430 431
                                 9, 9 /* FIXME */,
                                 mx >> 3, my >> 3,
                                 pic_width >> 1, pic_height >> 1);
        src_cb = h->edge_emu_buffer;
432
    }
433 434 435 436
    chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx & 7, my & 7);

    if (emu) {
        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
437
                                 h->c_stride, h->c_stride,
438 439 440 441
                                 9, 9 /* FIXME */,
                                 mx >> 3, my >> 3,
                                 pic_width >> 1, pic_height >> 1);
        src_cr = h->edge_emu_buffer;
442
    }
443
    chroma_op(dest_cr, src_cr, h->c_stride, chroma_height, mx & 7, my & 7);
444 445
}

446 447 448 449 450 451 452 453 454 455
static inline void mc_part_std(AVSContext *h, int chroma_height, int delta,
                               uint8_t *dest_y,
                               uint8_t *dest_cb,
                               uint8_t *dest_cr,
                               int x_offset, int y_offset,
                               qpel_mc_func *qpix_put,
                               h264_chroma_mc_func chroma_put,
                               qpel_mc_func *qpix_avg,
                               h264_chroma_mc_func chroma_avg,
                               cavs_vector *mv)
Aneesh Dogra's avatar
Aneesh Dogra committed
456
{
457 458
    qpel_mc_func *qpix_op =  qpix_put;
    h264_chroma_mc_func chroma_op = chroma_put;
459

460 461 462
    dest_y   += x_offset * 2 + y_offset * h->l_stride * 2;
    dest_cb  += x_offset     + y_offset * h->c_stride;
    dest_cr  += x_offset     + y_offset * h->c_stride;
463 464
    x_offset += 8 * h->mbx;
    y_offset += 8 * h->mby;
465

466
    if (mv->ref >= 0) {
467
        AVFrame *ref = h->DPB[mv->ref].f;
Aneesh Dogra's avatar
Aneesh Dogra committed
468
        mc_dir_part(h, ref, chroma_height, delta, 0,
469 470 471
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
                    qpix_op, chroma_op, mv);

472 473
        qpix_op   = qpix_avg;
        chroma_op = chroma_avg;
474 475
    }

476
    if ((mv + MV_BWD_OFFS)->ref >= 0) {
477
        AVFrame *ref = h->DPB[0].f;
Aneesh Dogra's avatar
Aneesh Dogra committed
478
        mc_dir_part(h, ref, chroma_height, delta, 1,
479
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
480
                    qpix_op, chroma_op, mv + MV_BWD_OFFS);
481 482 483
    }
}

484 485 486
void ff_cavs_inter(AVSContext *h, enum cavs_mb mb_type)
{
    if (ff_cavs_partition_flags[mb_type] == 0) { // 16x16
Aneesh Dogra's avatar
Aneesh Dogra committed
487
        mc_part_std(h, 8, 0, h->cy, h->cu, h->cv, 0, 0,
488 489 490 491 492 493
                    h->cdsp.put_cavs_qpel_pixels_tab[0],
                    h->h264chroma.put_h264_chroma_pixels_tab[0],
                    h->cdsp.avg_cavs_qpel_pixels_tab[0],
                    h->h264chroma.avg_h264_chroma_pixels_tab[0],
                    &h->mv[MV_FWD_X0]);
    } else {
Aneesh Dogra's avatar
Aneesh Dogra committed
494
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 0,
495 496 497 498 499
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
                    &h->mv[MV_FWD_X0]);
Aneesh Dogra's avatar
Aneesh Dogra committed
500
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 0,
501 502 503 504 505
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
                    &h->mv[MV_FWD_X1]);
Aneesh Dogra's avatar
Aneesh Dogra committed
506
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 0, 4,
507 508 509 510 511
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
                    &h->mv[MV_FWD_X2]);
Aneesh Dogra's avatar
Aneesh Dogra committed
512
        mc_part_std(h, 4, 0, h->cy, h->cu, h->cv, 4, 4,
513 514 515 516 517
                    h->cdsp.put_cavs_qpel_pixels_tab[1],
                    h->h264chroma.put_h264_chroma_pixels_tab[1],
                    h->cdsp.avg_cavs_qpel_pixels_tab[1],
                    h->h264chroma.avg_h264_chroma_pixels_tab[1],
                    &h->mv[MV_FWD_X3]);
518
    }
519 520 521 522 523 524 525 526
}

/*****************************************************************************
 *
 * motion vector prediction
 *
 ****************************************************************************/

527 528 529
static inline void scale_mv(AVSContext *h, int *d_x, int *d_y,
                            cavs_vector *src, int distp)
{
530 531
    int den = h->scale_den[src->ref];

532 533
    *d_x = (src->x * distp * den + 256 + FF_SIGNBIT(src->x)) >> 9;
    *d_y = (src->y * distp * den + 256 + FF_SIGNBIT(src->y)) >> 9;
534 535
}

536 537 538 539 540 541
static inline void mv_pred_median(AVSContext *h,
                                  cavs_vector *mvP,
                                  cavs_vector *mvA,
                                  cavs_vector *mvB,
                                  cavs_vector *mvC)
{
542 543 544 545 546 547 548 549
    int ax, ay, bx, by, cx, cy;
    int len_ab, len_bc, len_ca, len_mid;

    /* scale candidates according to their temporal span */
    scale_mv(h, &ax, &ay, mvA, mvP->dist);
    scale_mv(h, &bx, &by, mvB, mvP->dist);
    scale_mv(h, &cx, &cy, mvC, mvP->dist);
    /* find the geometrical median of the three candidates */
550 551 552
    len_ab  = abs(ax - bx) + abs(ay - by);
    len_bc  = abs(bx - cx) + abs(by - cy);
    len_ca  = abs(cx - ax) + abs(cy - ay);
553
    len_mid = mid_pred(len_ab, len_bc, len_ca);
554
    if (len_mid == len_ab) {
555 556
        mvP->x = cx;
        mvP->y = cy;
557
    } else if (len_mid == len_bc) {
558 559 560 561 562 563 564 565
        mvP->x = ax;
        mvP->y = ay;
    } else {
        mvP->x = bx;
        mvP->y = by;
    }
}

566
void ff_cavs_mv(AVSContext *h, enum cavs_mv_loc nP, enum cavs_mv_loc nC,
567 568
                enum cavs_mv_pred mode, enum cavs_block size, int ref)
{
569 570 571 572 573
    cavs_vector *mvP = &h->mv[nP];
    cavs_vector *mvA = &h->mv[nP-1];
    cavs_vector *mvB = &h->mv[nP-4];
    cavs_vector *mvC = &h->mv[nC];
    const cavs_vector *mvP2 = NULL;
574

575
    mvP->ref  = ref;
576
    mvP->dist = h->dist[mvP->ref];
577
    if (mvC->ref == NOT_AVAIL)
578
        mvC = &h->mv[nP - 5];  // set to top-left (mvD)
579 580 581 582 583
    if (mode == MV_PRED_PSKIP &&
        (mvA->ref == NOT_AVAIL ||
         mvB->ref == NOT_AVAIL ||
         (mvA->x | mvA->y | mvA->ref) == 0 ||
         (mvB->x | mvB->y | mvB->ref) == 0)) {
584
        mvP2 = &un_mv;
585
    /* if there is only one suitable candidate, take it */
586 587 588 589 590 591 592 593 594 595 596 597
    } else if (mvA->ref >= 0 && mvB->ref < 0  && mvC->ref < 0) {
        mvP2 = mvA;
    } else if (mvA->ref < 0  && mvB->ref >= 0 && mvC->ref < 0) {
        mvP2 = mvB;
    } else if (mvA->ref < 0  && mvB->ref < 0  && mvC->ref >= 0) {
        mvP2 = mvC;
    } else if (mode == MV_PRED_LEFT     && mvA->ref == ref) {
        mvP2 = mvA;
    } else if (mode == MV_PRED_TOP      && mvB->ref == ref) {
        mvP2 = mvB;
    } else if (mode == MV_PRED_TOPRIGHT && mvC->ref == ref) {
        mvP2 = mvC;
598
    }
599
    if (mvP2) {
Michael Niedermayer's avatar
Michael Niedermayer committed
600 601
        mvP->x = mvP2->x;
        mvP->y = mvP2->y;
602
    } else
Michael Niedermayer's avatar
Michael Niedermayer committed
603 604
        mv_pred_median(h, mvP, mvA, mvB, mvC);

605
    if (mode < MV_PRED_PSKIP) {
606 607
        mvP->x += get_se_golomb(&h->bc);
        mvP->y += get_se_golomb(&h->bc);
608
    }
609
    set_mvs(mvP, size);
610 611
}

612 613 614 615 616 617 618 619 620
/*****************************************************************************
 *
 * macroblock level
 *
 ****************************************************************************/

/**
 * initialise predictors for motion vectors and intra prediction
 */
621 622
void ff_cavs_init_mb(AVSContext *h)
{
623 624 625
    int i;

    /* copy predictors from top line (MB B and C) into cache */
626 627 628
    for (i = 0; i < 3; i++) {
        h->mv[MV_FWD_B2 + i] = h->top_mv[0][h->mbx * 2 + i];
        h->mv[MV_BWD_B2 + i] = h->top_mv[1][h->mbx * 2 + i];
629
    }
630 631
    h->pred_mode_Y[1] = h->top_pred_Y[h->mbx * 2 + 0];
    h->pred_mode_Y[2] = h->top_pred_Y[h->mbx * 2 + 1];
632
    /* clear top predictors if MB B is not available */
633 634 635 636 637
    if (!(h->flags & B_AVAIL)) {
        h->mv[MV_FWD_B2]  = un_mv;
        h->mv[MV_FWD_B3]  = un_mv;
        h->mv[MV_BWD_B2]  = un_mv;
        h->mv[MV_BWD_B3]  = un_mv;
638
        h->pred_mode_Y[1] = h->pred_mode_Y[2] = NOT_AVAIL;
639 640
        h->flags         &= ~(C_AVAIL | D_AVAIL);
    } else if (h->mbx) {
641 642
        h->flags |= D_AVAIL;
    }
643
    if (h->mbx == h->mb_width - 1) // MB C not available
644 645
        h->flags &= ~C_AVAIL;
    /* clear top-right predictors if MB C is not available */
646
    if (!(h->flags & C_AVAIL)) {
647 648
        h->mv[MV_FWD_C2] = un_mv;
        h->mv[MV_BWD_C2] = un_mv;
649 650
    }
    /* clear top-left predictors if MB D is not available */
651
    if (!(h->flags & D_AVAIL)) {
652 653
        h->mv[MV_FWD_D3] = un_mv;
        h->mv[MV_BWD_D3] = un_mv;
654 655 656 657 658 659
    }
}

/**
 * save predictors for later macroblocks and increase
 * macroblock address
660
 * @return 0 if end of frame is reached, 1 otherwise
661
 */
662 663
int ff_cavs_next_mb(AVSContext *h)
{
664 665 666
    int i;

    h->flags |= A_AVAIL;
667 668 669
    h->cy    += 16;
    h->cu    += 8;
    h->cv    += 8;
670
    /* copy mvs as predictors to the left */
671 672
    for (i = 0; i <= 20; i += 4)
        h->mv[i] = h->mv[i + 2];
673
    /* copy bottom mvs from cache to top line */
674 675 676 677
    h->top_mv[0][h->mbx * 2 + 0] = h->mv[MV_FWD_X2];
    h->top_mv[0][h->mbx * 2 + 1] = h->mv[MV_FWD_X3];
    h->top_mv[1][h->mbx * 2 + 0] = h->mv[MV_BWD_X2];
    h->top_mv[1][h->mbx * 2 + 1] = h->mv[MV_BWD_X3];
678
    /* next MB address */
679
    h->mbidx++;
680
    h->mbx++;
681 682
    if (h->mbx == h->mb_width) { // New mb line
        h->flags = B_AVAIL | C_AVAIL;
683 684 685
        /* clear left pred_modes */
        h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
        /* clear left mv predictors */
686
        for (i = 0; i <= 20; i += 4)
687
            h->mv[i] = un_mv;
688 689 690
        h->mbx = 0;
        h->mby++;
        /* re-calculate sample pointers */
691
        h->cy = h->cur.f->data[0] + h->mby * 16 * h->l_stride;
692 693 694
        h->cu = h->cur.f->data[1] + h->mby * 8 * h->c_stride;
        h->cv = h->cur.f->data[2] + h->mby * 8 * h->c_stride;
        if (h->mby == h->mb_height) { // Frame end
695 696 697 698 699 700
            return 0;
        }
    }
    return 1;
}

701 702 703 704 705 706
/*****************************************************************************
 *
 * frame level
 *
 ****************************************************************************/

707 708
void ff_cavs_init_pic(AVSContext *h)
{
709 710 711
    int i;

    /* clear some predictors */
712
    for (i = 0; i <= 20; i += 4)
713
        h->mv[i] = un_mv;
714
    h->mv[MV_BWD_X0] = ff_cavs_dir_mv;
715
    set_mvs(&h->mv[MV_BWD_X0], BLK_16X16);
716
    h->mv[MV_FWD_X0] = ff_cavs_dir_mv;
717 718
    set_mvs(&h->mv[MV_FWD_X0], BLK_16X16);
    h->pred_mode_Y[3] = h->pred_mode_Y[6] = NOT_AVAIL;
719 720 721 722 723 724 725 726 727
    h->cy             = h->cur.f->data[0];
    h->cu             = h->cur.f->data[1];
    h->cv             = h->cur.f->data[2];
    h->l_stride       = h->cur.f->linesize[0];
    h->c_stride       = h->cur.f->linesize[1];
    h->luma_scan[2]   = 8 * h->l_stride;
    h->luma_scan[3]   = 8 * h->l_stride + 8;
    h->mbx            = h->mby = h->mbidx = 0;
    h->flags          = 0;
728 729 730 731 732 733 734 735
}

/*****************************************************************************
 *
 * headers and interface
 *
 ****************************************************************************/

Stefan Gehrer's avatar
Stefan Gehrer committed
736 737 738 739 740
/**
 * some predictions require data from the top-neighbouring macroblock.
 * this data has to be stored for one complete row of macroblocks
 * and this storage space is allocated here
 */
741 742
void ff_cavs_init_top_lines(AVSContext *h)
{
743
    /* alloc top line of predictors */
744 745 746 747 748 749 750
    h->top_qp       = av_mallocz(h->mb_width);
    h->top_mv[0]    = av_mallocz((h->mb_width * 2 + 1) * sizeof(cavs_vector));
    h->top_mv[1]    = av_mallocz((h->mb_width * 2 + 1) * sizeof(cavs_vector));
    h->top_pred_Y   = av_mallocz(h->mb_width * 2 * sizeof(*h->top_pred_Y));
    h->top_border_y = av_mallocz((h->mb_width + 1) * 16);
    h->top_border_u = av_mallocz(h->mb_width * 10);
    h->top_border_v = av_mallocz(h->mb_width * 10);
751 752

    /* alloc space for co-located MVs and types */
753 754 755 756
    h->col_mv        = av_mallocz(h->mb_width * h->mb_height * 4 *
                                  sizeof(cavs_vector));
    h->col_type_base = av_mallocz(h->mb_width * h->mb_height);
    h->block         = av_mallocz(64 * sizeof(int16_t));
757 758
}

759 760
av_cold int ff_cavs_init(AVCodecContext *avctx)
{
Måns Rullgård's avatar
Måns Rullgård committed
761
    AVSContext *h = avctx->priv_data;
762

763
    ff_blockdsp_init(&h->bdsp, avctx);
764
    ff_h264chroma_init(&h->h264chroma, 8);
765
    ff_idctdsp_init(&h->idsp, avctx);
766
    ff_videodsp_init(&h->vdsp, 8);
767
    ff_cavsdsp_init(&h->cdsp, avctx);
768
    ff_init_scantable_permutation(h->idsp.idct_permutation,
769
                                  h->cdsp.idct_perm);
770
    ff_init_scantable(h->idsp.idct_permutation, &h->scantable, ff_zigzag_direct);
771

772 773
    h->avctx       = avctx;
    avctx->pix_fmt = AV_PIX_FMT_YUV420P;
774

775 776 777
    h->cur.f    = av_frame_alloc();
    h->DPB[0].f = av_frame_alloc();
    h->DPB[1].f = av_frame_alloc();
778 779 780 781 782
    if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f) {
        ff_cavs_end(avctx);
        return AVERROR(ENOMEM);
    }

783 784 785 786 787 788
    h->luma_scan[0]                     = 0;
    h->luma_scan[1]                     = 8;
    h->intra_pred_l[INTRA_L_VERT]       = intra_pred_vert;
    h->intra_pred_l[INTRA_L_HORIZ]      = intra_pred_horiz;
    h->intra_pred_l[INTRA_L_LP]         = intra_pred_lp;
    h->intra_pred_l[INTRA_L_DOWN_LEFT]  = intra_pred_down_left;
789
    h->intra_pred_l[INTRA_L_DOWN_RIGHT] = intra_pred_down_right;
790 791 792 793 794 795 796 797 798 799 800 801
    h->intra_pred_l[INTRA_L_LP_LEFT]    = intra_pred_lp_left;
    h->intra_pred_l[INTRA_L_LP_TOP]     = intra_pred_lp_top;
    h->intra_pred_l[INTRA_L_DC_128]     = intra_pred_dc_128;
    h->intra_pred_c[INTRA_C_LP]         = intra_pred_lp;
    h->intra_pred_c[INTRA_C_HORIZ]      = intra_pred_horiz;
    h->intra_pred_c[INTRA_C_VERT]       = intra_pred_vert;
    h->intra_pred_c[INTRA_C_PLANE]      = intra_pred_plane;
    h->intra_pred_c[INTRA_C_LP_LEFT]    = intra_pred_lp_left;
    h->intra_pred_c[INTRA_C_LP_TOP]     = intra_pred_lp_top;
    h->intra_pred_c[INTRA_C_DC_128]     = intra_pred_dc_128;
    h->mv[7]                            = un_mv;
    h->mv[19]                           = un_mv;
802 803 804
    return 0;
}

805 806
av_cold int ff_cavs_end(AVCodecContext *avctx)
{
Måns Rullgård's avatar
Måns Rullgård committed
807
    AVSContext *h = avctx->priv_data;
808

809 810 811
    av_frame_free(&h->cur.f);
    av_frame_free(&h->DPB[0].f);
    av_frame_free(&h->DPB[1].f);
812

813 814 815 816 817 818 819 820 821
    av_free(h->top_qp);
    av_free(h->top_mv[0]);
    av_free(h->top_mv[1]);
    av_free(h->top_pred_Y);
    av_free(h->top_border_y);
    av_free(h->top_border_u);
    av_free(h->top_border_v);
    av_free(h->col_mv);
    av_free(h->col_type_base);
822
    av_free(h->block);
823
    av_freep(&h->edge_emu_buffer);
824 825
    return 0;
}