svq1enc.c 23.6 KB
Newer Older
1 2 3 4
/*
 * SVQ1 Encoder
 * Copyright (C) 2004 Mike Melanson <melanson@pcisys.net>
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9 10 11
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
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 20 21 22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
23
 * @file
24 25 26 27 28 29
 * Sorenson Vector Quantizer #1 (SVQ1) video codec.
 * For more information of the SVQ1 algorithm, visit:
 *   http://www.pcisys.net/~melanson/codecs/
 */

#include "avcodec.h"
30
#include "hpeldsp.h"
31
#include "me_cmp.h"
32
#include "mpegvideo.h"
33
#include "h263.h"
34
#include "internal.h"
35
#include "mpegutils.h"
36
#include "svq1.h"
37
#include "svq1enc.h"
38 39 40 41 42
#include "svq1enc_cb.h"

#undef NDEBUG
#include <assert.h>

43
static void svq1_write_header(SVQ1EncContext *s, int frame_type)
44 45 46 47 48 49 50 51 52 53 54 55
{
    int i;

    /* frame code */
    put_bits(&s->pb, 22, 0x20);

    /* temporal reference (sure hope this is a "don't care") */
    put_bits(&s->pb, 8, 0x00);

    /* frame type */
    put_bits(&s->pb, 2, frame_type - 1);

56
    if (frame_type == AV_PICTURE_TYPE_I) {
57 58 59 60 61
        /* no checksum since frame code is 0x20 */
        /* no embedded string either */
        /* output 5 unknown bits (2 + 2 + 1) */
        put_bits(&s->pb, 5, 2); /* 2 needed by quicktime decoder */

62 63 64
        i = ff_match_2uint16(ff_svq1_frame_size_table,
                             FF_ARRAY_ELEMS(ff_svq1_frame_size_table),
                             s->frame_width, s->frame_height);
65
        put_bits(&s->pb, 3, i);
66

67 68 69
        if (i == 7) {
            put_bits(&s->pb, 12, s->frame_width);
            put_bits(&s->pb, 12, s->frame_height);
70 71 72 73 74 75 76
        }
    }

    /* no checksum or extra data (next 2 bits get 0) */
    put_bits(&s->pb, 2, 0);
}

77
#define QUALITY_THRESHOLD    100
78 79
#define THRESHOLD_MULTIPLIER 0.6

80 81 82 83 84 85 86 87 88 89
static int ssd_int8_vs_int16_c(const int8_t *pix1, const int16_t *pix2,
                               int size)
{
    int score = 0, i;

    for (i = 0; i < size; i++)
        score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
    return score;
}

90
static int encode_block(SVQ1EncContext *s, uint8_t *src, uint8_t *ref,
91 92 93
                        uint8_t *decoded, int stride, int level,
                        int threshold, int lambda, int intra)
{
94 95
    int count, y, x, i, j, split, best_mean, best_score, best_count;
    int best_vector[6];
96
    int block_sum[7] = { 0, 0, 0, 0, 0, 0 };
97 98
    int w            = 2 << (level + 2 >> 1);
    int h            = 2 << (level + 1 >> 1);
99
    int size         = w * h;
100 101
    int16_t block[7][256];
    const int8_t *codebook_sum, *codebook;
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
    const uint16_t(*mean_vlc)[2];
    const uint8_t(*multistage_vlc)[2];

    best_score = 0;
    // FIXME: Optimize, this does not need to be done multiple times.
    if (intra) {
        codebook_sum   = svq1_intra_codebook_sum[level];
        codebook       = ff_svq1_intra_codebooks[level];
        mean_vlc       = ff_svq1_intra_mean_vlc;
        multistage_vlc = ff_svq1_intra_multistage_vlc[level];
        for (y = 0; y < h; y++) {
            for (x = 0; x < w; x++) {
                int v = src[x + y * stride];
                block[0][x + w * y] = v;
                best_score         += v * v;
                block_sum[0]       += v;
118 119
            }
        }
120 121 122 123 124 125 126 127 128 129 130
    } else {
        codebook_sum   = svq1_inter_codebook_sum[level];
        codebook       = ff_svq1_inter_codebooks[level];
        mean_vlc       = ff_svq1_inter_mean_vlc + 256;
        multistage_vlc = ff_svq1_inter_multistage_vlc[level];
        for (y = 0; y < h; y++) {
            for (x = 0; x < w; x++) {
                int v = src[x + y * stride] - ref[x + y * stride];
                block[0][x + w * y] = v;
                best_score         += v * v;
                block_sum[0]       += v;
131 132 133 134
            }
        }
    }

135
    best_count  = 0;
136 137
    best_score -= (int)((unsigned)block_sum[0] * block_sum[0] >> (level + 3));
    best_mean   = block_sum[0] + (size >> 1) >> (level + 3);
138

139 140 141 142 143
    if (level < 4) {
        for (count = 1; count < 7; count++) {
            int best_vector_score = INT_MAX;
            int best_vector_sum   = -999, best_vector_mean = -999;
            const int stage       = count - 1;
144 145
            const int8_t *vector;

146 147
            for (i = 0; i < 16; i++) {
                int sum = codebook_sum[stage * 16 + i];
148 149
                int sqr, diff, score;

150
                vector = codebook + stage * size * 16 + i * size;
151
                sqr    = s->ssd_int8_vs_int16(vector, block[stage], size);
152
                diff   = block_sum[stage] - sum;
153
                score  = sqr - (diff * (int64_t)diff >> (level + 3)); // FIXME: 64bit slooow
154
                if (score < best_vector_score) {
155
                    int mean = diff + (size >> 1) >> (level + 3);
156 157 158 159 160 161
                    assert(mean > -300 && mean < 300);
                    mean               = av_clip(mean, intra ? 0 : -256, 255);
                    best_vector_score  = score;
                    best_vector[stage] = i;
                    best_vector_sum    = sum;
                    best_vector_mean   = mean;
162 163 164
                }
            }
            assert(best_vector_mean != -999);
165 166 167 168 169 170 171 172 173 174 175 176 177
            vector = codebook + stage * size * 16 + best_vector[stage] * size;
            for (j = 0; j < size; j++)
                block[stage + 1][j] = block[stage][j] - vector[j];
            block_sum[stage + 1] = block_sum[stage] - best_vector_sum;
            best_vector_score   += lambda *
                                   (+1 + 4 * count +
                                    multistage_vlc[1 + count][1]
                                    + mean_vlc[best_vector_mean][1]);

            if (best_vector_score < best_score) {
                best_score = best_vector_score;
                best_count = count;
                best_mean  = best_vector_mean;
178 179 180 181
            }
        }
    }

182 183 184
    split = 0;
    if (best_score > threshold && level) {
        int score  = 0;
185
        int offset = level & 1 ? stride * h / 2 : w / 2;
186 187
        PutBitContext backup[6];

188 189 190 191 192 193
        for (i = level - 1; i >= 0; i--)
            backup[i] = s->reorder_pb[i];
        score += encode_block(s, src, ref, decoded, stride, level - 1,
                              threshold >> 1, lambda, intra);
        score += encode_block(s, src + offset, ref + offset, decoded + offset,
                              stride, level - 1, threshold >> 1, lambda, intra);
194 195
        score += lambda;

196 197 198 199 200 201
        if (score < best_score) {
            best_score = score;
            split      = 1;
        } else {
            for (i = level - 1; i >= 0; i--)
                s->reorder_pb[i] = backup[i];
202 203 204 205 206
        }
    }
    if (level > 0)
        put_bits(&s->reorder_pb[level], 1, split);

207
    if (!split) {
208
        assert(best_mean >= 0 && best_mean < 256 || !intra);
209 210 211
        assert(best_mean >= -256 && best_mean < 256);
        assert(best_count >= 0 && best_count < 7);
        assert(level < 4 || best_count == 0);
212 213 214

        /* output the encoding */
        put_bits(&s->reorder_pb[level],
215 216
                 multistage_vlc[1 + best_count][1],
                 multistage_vlc[1 + best_count][0]);
217
        put_bits(&s->reorder_pb[level], mean_vlc[best_mean][1],
218
                 mean_vlc[best_mean][0]);
219

220 221
        for (i = 0; i < best_count; i++) {
            assert(best_vector[i] >= 0 && best_vector[i] < 16);
222 223 224
            put_bits(&s->reorder_pb[level], 4, best_vector[i]);
        }

225 226 227 228 229
        for (y = 0; y < h; y++)
            for (x = 0; x < w; x++)
                decoded[x + y * stride] = src[x + y * stride] -
                                          block[best_count][x + w * y] +
                                          best_mean;
230 231 232 233 234
    }

    return best_score;
}

235
static int svq1_encode_plane(SVQ1EncContext *s, int plane,
236 237 238 239
                             unsigned char *src_plane,
                             unsigned char *ref_plane,
                             unsigned char *decoded_plane,
                             int width, int height, int src_stride, int stride)
240
{
241
    const AVFrame *f = s->avctx->coded_frame;
242 243 244 245 246
    int x, y;
    int i;
    int block_width, block_height;
    int level;
    int threshold[6];
247
    uint8_t *src     = s->scratchbuf + stride * 16;
248
    const int lambda = (f->quality * f->quality) >>
249
                       (2 * FF_LAMBDA_SHIFT);
250 251 252 253 254 255

    /* figure out the acceptable level thresholds in advance */
    threshold[5] = QUALITY_THRESHOLD;
    for (level = 4; level >= 0; level--)
        threshold[level] = threshold[level + 1] * THRESHOLD_MULTIPLIER;

256
    block_width  = (width  + 15) / 16;
257 258
    block_height = (height + 15) / 16;

259
    if (f->pict_type == AV_PICTURE_TYPE_P) {
260 261 262
        s->m.avctx                         = s->avctx;
        s->m.current_picture_ptr           = &s->m.current_picture;
        s->m.last_picture_ptr              = &s->m.last_picture;
263
        s->m.last_picture.f->data[0]        = ref_plane;
264
        s->m.linesize                      =
265 266 267
        s->m.last_picture.f->linesize[0]    =
        s->m.new_picture.f->linesize[0]     =
        s->m.current_picture.f->linesize[0] = stride;
268 269 270 271 272 273 274
        s->m.width                         = width;
        s->m.height                        = height;
        s->m.mb_width                      = block_width;
        s->m.mb_height                     = block_height;
        s->m.mb_stride                     = s->m.mb_width + 1;
        s->m.b8_stride                     = 2 * s->m.mb_width + 1;
        s->m.f_code                        = 1;
275
        s->m.pict_type                     = f->pict_type;
276 277 278 279 280
        s->m.me_method                     = s->avctx->me_method;
        s->m.me.scene_change_score         = 0;
        s->m.flags                         = s->avctx->flags;
        // s->m.out_format                    = FMT_H263;
        // s->m.unrestricted_mv               = 1;
281
        s->m.lambda                        = f->quality;
282 283 284 285 286
        s->m.qscale                        = s->m.lambda * 139 +
                                             FF_LAMBDA_SCALE * 64 >>
                                             FF_LAMBDA_SHIFT + 7;
        s->m.lambda2                       = s->m.lambda * s->m.lambda +
                                             FF_LAMBDA_SCALE / 2 >>
287 288 289 290 291 292 293 294 295
                                             FF_LAMBDA_SHIFT;

        if (!s->motion_val8[plane]) {
            s->motion_val8[plane]  = av_mallocz((s->m.b8_stride *
                                                 block_height * 2 + 2) *
                                                2 * sizeof(int16_t));
            s->motion_val16[plane] = av_mallocz((s->m.mb_stride *
                                                 (block_height + 2) + 1) *
                                                2 * sizeof(int16_t));
296 297
        }

298
        s->m.mb_type = s->mb_type;
299

300 301 302 303
        // dummies, to avoid segfaults
        s->m.current_picture.mb_mean   = (uint8_t *)s->dummy;
        s->m.current_picture.mb_var    = (uint16_t *)s->dummy;
        s->m.current_picture.mc_mb_var = (uint16_t *)s->dummy;
304
        s->m.current_picture.mb_type = s->dummy;
305

306
        s->m.current_picture.motion_val[0]   = s->motion_val8[plane] + 2;
307 308
        s->m.p_mv_table                      = s->motion_val16[plane] +
                                               s->m.mb_stride + 1;
309
        s->m.mecc                            = s->mecc; // move
310 311
        ff_init_me(&s->m);

312 313
        s->m.me.dia_size      = s->avctx->dia_size;
        s->m.first_slice_line = 1;
314
        for (y = 0; y < block_height; y++) {
315
            s->m.new_picture.f->data[0] = src - y * 16 * stride; // ugly
316 317 318 319 320 321 322
            s->m.mb_y                  = y;

            for (i = 0; i < 16 && i + 16 * y < height; i++) {
                memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
                       width);
                for (x = width; x < 16 * block_width; x++)
                    src[i * stride + x] = src[i * stride + x - 1];
323
            }
324 325 326
            for (; i < 16 && i + 16 * y < 16 * block_height; i++)
                memcpy(&src[i * stride], &src[(i - 1) * stride],
                       16 * block_width);
327 328

            for (x = 0; x < block_width; x++) {
329
                s->m.mb_x = x;
330 331 332 333 334
                ff_init_block_index(&s->m);
                ff_update_block_index(&s->m);

                ff_estimate_p_frame_motion(&s->m, x, y);
            }
335
            s->m.first_slice_line = 0;
336 337 338
        }

        ff_fix_long_p_mvs(&s->m);
339 340
        ff_fix_long_mvs(&s->m, NULL, 0, s->m.p_mv_table, s->m.f_code,
                        CANDIDATE_MB_TYPE_INTER, 0);
341 342
    }

343
    s->m.first_slice_line = 1;
344
    for (y = 0; y < block_height; y++) {
345 346 347 348 349
        for (i = 0; i < 16 && i + 16 * y < height; i++) {
            memcpy(&src[i * stride], &src_plane[(i + 16 * y) * src_stride],
                   width);
            for (x = width; x < 16 * block_width; x++)
                src[i * stride + x] = src[i * stride + x - 1];
350
        }
351 352
        for (; i < 16 && i + 16 * y < 16 * block_height; i++)
            memcpy(&src[i * stride], &src[(i - 1) * stride], 16 * block_width);
353

354
        s->m.mb_y = y;
355
        for (x = 0; x < block_width; x++) {
356
            uint8_t reorder_buffer[3][6][7 * 32];
357
            int count[3][6];
358 359 360 361 362 363 364 365
            int offset       = y * 16 * stride + x * 16;
            uint8_t *decoded = decoded_plane + offset;
            uint8_t *ref     = ref_plane + offset;
            int score[4]     = { 0, 0, 0, 0 }, best;
            uint8_t *temp    = s->scratchbuf;

            if (s->pb.buf_end - s->pb.buf -
                (put_bits_count(&s->pb) >> 3) < 3000) { // FIXME: check size
366 367 368 369
                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
                return -1;
            }

370
            s->m.mb_x = x;
371 372 373
            ff_init_block_index(&s->m);
            ff_update_block_index(&s->m);

374
            if (f->pict_type == AV_PICTURE_TYPE_I ||
375 376 377 378 379
                (s->m.mb_type[x + y * s->m.mb_stride] &
                 CANDIDATE_MB_TYPE_INTRA)) {
                for (i = 0; i < 6; i++)
                    init_put_bits(&s->reorder_pb[i], reorder_buffer[0][i],
                                  7 * 32);
380
                if (f->pict_type == AV_PICTURE_TYPE_P) {
381
                    const uint8_t *vlc = ff_svq1_block_type_vlc[SVQ1_BLOCK_INTRA];
382
                    put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);
383
                    score[0] = vlc[1] * lambda;
384
                }
385 386 387 388
                score[0] += encode_block(s, src + 16 * x, NULL, temp, stride,
                                         5, 64, lambda, 1);
                for (i = 0; i < 6; i++) {
                    count[0][i] = put_bits_count(&s->reorder_pb[i]);
389 390
                    flush_put_bits(&s->reorder_pb[i]);
                }
391 392
            } else
                score[0] = INT_MAX;
393

394
            best = 0;
395

396
            if (f->pict_type == AV_PICTURE_TYPE_P) {
397
                const uint8_t *vlc = ff_svq1_block_type_vlc[SVQ1_BLOCK_INTER];
398 399 400
                int mx, my, pred_x, pred_y, dxy;
                int16_t *motion_ptr;

401 402 403 404 405 406
                motion_ptr = ff_h263_pred_motion(&s->m, 0, 0, &pred_x, &pred_y);
                if (s->m.mb_type[x + y * s->m.mb_stride] &
                    CANDIDATE_MB_TYPE_INTER) {
                    for (i = 0; i < 6; i++)
                        init_put_bits(&s->reorder_pb[i], reorder_buffer[1][i],
                                      7 * 32);
407 408 409

                    put_bits(&s->reorder_pb[5], vlc[1], vlc[0]);

410 411 412 413 414 415 416
                    s->m.pb = s->reorder_pb[5];
                    mx      = motion_ptr[0];
                    my      = motion_ptr[1];
                    assert(mx     >= -32 && mx     <= 31);
                    assert(my     >= -32 && my     <= 31);
                    assert(pred_x >= -32 && pred_x <= 31);
                    assert(pred_y >= -32 && pred_y <= 31);
417 418
                    ff_h263_encode_motion(&s->m, mx - pred_x, 1);
                    ff_h263_encode_motion(&s->m, my - pred_y, 1);
419 420 421 422 423
                    s->reorder_pb[5] = s->m.pb;
                    score[1]        += lambda * put_bits_count(&s->reorder_pb[5]);

                    dxy = (mx & 1) + 2 * (my & 1);

424 425 426 427
                    s->hdsp.put_pixels_tab[0][dxy](temp + 16,
                                                   ref + (mx >> 1) +
                                                   stride * (my >> 1),
                                                   stride, 16);
428 429 430 431 432 433

                    score[1] += encode_block(s, src + 16 * x, temp + 16,
                                             decoded, stride, 5, 64, lambda, 0);
                    best      = score[1] <= score[0];

                    vlc       = ff_svq1_block_type_vlc[SVQ1_BLOCK_SKIP];
434 435
                    score[2]  = s->mecc.sse[0](NULL, src + 16 * x, ref,
                                               stride, 16);
436 437 438
                    score[2] += vlc[1] * lambda;
                    if (score[2] < score[best] && mx == 0 && my == 0) {
                        best = 2;
439
                        s->hdsp.put_pixels_tab[0][0](decoded, ref, stride, 16);
440 441
                        for (i = 0; i < 6; i++)
                            count[2][i] = 0;
442 443 444 445
                        put_bits(&s->pb, vlc[1], vlc[0]);
                    }
                }

446 447 448
                if (best == 1) {
                    for (i = 0; i < 6; i++) {
                        count[1][i] = put_bits_count(&s->reorder_pb[i]);
449 450
                        flush_put_bits(&s->reorder_pb[i]);
                    }
451 452 453 454 455 456 457 458 459
                } else {
                    motion_ptr[0]                      =
                    motion_ptr[1]                      =
                    motion_ptr[2]                      =
                    motion_ptr[3]                      =
                    motion_ptr[0 + 2 * s->m.b8_stride] =
                    motion_ptr[1 + 2 * s->m.b8_stride] =
                    motion_ptr[2 + 2 * s->m.b8_stride] =
                    motion_ptr[3 + 2 * s->m.b8_stride] = 0;
460 461 462 463 464
                }
            }

            s->rd_total += score[best];

465 466 467 468
            for (i = 5; i >= 0; i--)
                avpriv_copy_bits(&s->pb, reorder_buffer[best][i],
                                 count[best][i]);
            if (best == 0)
469
                s->hdsp.put_pixels_tab[0][0](decoded, temp, stride, 16);
470
        }
471
        s->m.first_slice_line = 0;
472 473 474 475
    }
    return 0;
}

476 477
static av_cold int svq1_encode_end(AVCodecContext *avctx)
{
478
    SVQ1EncContext *const s = avctx->priv_data;
479 480 481 482 483 484
    int i;

    av_log(avctx, AV_LOG_DEBUG, "RD: %f\n",
           s->rd_total / (double)(avctx->width * avctx->height *
                                  avctx->frame_number));

485
    s->m.mb_type = NULL;
486
    ff_mpv_common_end(&s->m);
487

488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506
    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->mb_type);
    av_freep(&s->dummy);
    av_freep(&s->scratchbuf);

    for (i = 0; i < 3; i++) {
        av_freep(&s->motion_val8[i]);
        av_freep(&s->motion_val16[i]);
    }

    av_frame_free(&s->current_picture);
    av_frame_free(&s->last_picture);
    av_frame_free(&avctx->coded_frame);

    return 0;
}

507
static av_cold int svq1_encode_init(AVCodecContext *avctx)
508
{
509
    SVQ1EncContext *const s = avctx->priv_data;
510
    int ret;
511

512
    ff_hpeldsp_init(&s->hdsp, avctx->flags);
513
    ff_me_cmp_init(&s->mecc, avctx);
514
    ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx);
515 516 517 518 519 520 521 522

    avctx->coded_frame = av_frame_alloc();
    s->current_picture = av_frame_alloc();
    s->last_picture    = av_frame_alloc();
    if (!avctx->coded_frame || !s->current_picture || !s->last_picture) {
        svq1_encode_end(avctx);
        return AVERROR(ENOMEM);
    }
523

524
    s->frame_width  = avctx->width;
525 526
    s->frame_height = avctx->height;

527
    s->y_block_width  = (s->frame_width  + 15) / 16;
528 529
    s->y_block_height = (s->frame_height + 15) / 16;

530
    s->c_block_width  = (s->frame_width  / 4 + 15) / 16;
531 532
    s->c_block_height = (s->frame_height / 4 + 15) / 16;

533 534
    s->avctx               = avctx;
    s->m.avctx             = avctx;
535

536
    if ((ret = ff_mpv_common_init(&s->m)) < 0) {
537 538 539 540
        svq1_encode_end(avctx);
        return ret;
    }

541
    s->m.picture_structure = PICT_FRAME;
542 543 544 545 546 547 548 549 550
    s->m.me.temp           =
    s->m.me.scratchpad     = av_mallocz((avctx->width + 64) *
                                        2 * 16 * 2 * sizeof(uint8_t));
    s->m.me.map            = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
    s->m.me.score_map      = av_mallocz(ME_MAP_SIZE * sizeof(uint32_t));
    s->mb_type             = av_mallocz((s->y_block_width + 1) *
                                        s->y_block_height * sizeof(int16_t));
    s->dummy               = av_mallocz((s->y_block_width + 1) *
                                        s->y_block_height * sizeof(int32_t));
551 552 553 554 555 556 557
    s->ssd_int8_vs_int16   = ssd_int8_vs_int16_c;

    if (ARCH_PPC)
        ff_svq1enc_init_ppc(s);
    if (ARCH_X86)
        ff_svq1enc_init_x86(s);

558
    ff_h263_encode_init(&s->m); // mv_penalty
559 560 561 562

    return 0;
}

563 564
static int svq1_encode_frame(AVCodecContext *avctx, AVPacket *pkt,
                             const AVFrame *pict, int *got_packet)
565
{
566 567
    SVQ1EncContext *const s = avctx->priv_data;
    AVFrame *const p        = avctx->coded_frame;
568 569 570
    int i, ret;

    if (!pkt->data &&
571
        (ret = av_new_packet(pkt, s->y_block_width * s->y_block_height *
572
                             MAX_MB_BYTES * 3 + FF_MIN_BUFFER_SIZE)) < 0) {
573 574 575
        av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
        return ret;
    }
576

577
    if (avctx->pix_fmt != AV_PIX_FMT_YUV410P) {
578 579 580 581
        av_log(avctx, AV_LOG_ERROR, "unsupported pixel format\n");
        return -1;
    }

582 583 584 585
    if (!s->current_picture->data[0]) {
        ff_get_buffer(avctx, s->current_picture, 0);
        ff_get_buffer(avctx, s->last_picture, 0);
        s->scratchbuf = av_malloc(s->current_picture->linesize[0] * 16 * 2);
586 587
    }

588
    FFSWAP(AVFrame*, s->current_picture, s->last_picture);
589

590
    init_put_bits(&s->pb, pkt->data, pkt->size);
591

592 593
    p->pict_type = avctx->gop_size && avctx->frame_number % avctx->gop_size ?
                   AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_I;
594
    p->key_frame = p->pict_type == AV_PICTURE_TYPE_I;
595
    p->quality   = pict->quality;
596 597

    svq1_write_header(s, p->pict_type);
598 599
    for (i = 0; i < 3; i++)
        if (svq1_encode_plane(s, i,
600 601 602
                              pict->data[i],
                              s->last_picture->data[i],
                              s->current_picture->data[i],
603 604
                              s->frame_width  / (i ? 4 : 1),
                              s->frame_height / (i ? 4 : 1),
605 606
                              pict->linesize[i],
                              s->current_picture->linesize[i]) < 0)
607 608 609 610
            return -1;

    // avpriv_align_put_bits(&s->pb);
    while (put_bits_count(&s->pb) & 31)
611 612 613 614
        put_bits(&s->pb, 1, 0);

    flush_put_bits(&s->pb);

615 616 617 618 619 620
    pkt->size = put_bits_count(&s->pb) / 8;
    if (p->pict_type == AV_PICTURE_TYPE_I)
        pkt->flags |= AV_PKT_FLAG_KEY;
    *got_packet = 1;

    return 0;
621 622
}

623
AVCodec ff_svq1_encoder = {
624
    .name           = "svq1",
625
    .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 1 / Sorenson Video 1 / SVQ1"),
626
    .type           = AVMEDIA_TYPE_VIDEO,
627
    .id             = AV_CODEC_ID_SVQ1,
628
    .priv_data_size = sizeof(SVQ1EncContext),
629
    .init           = svq1_encode_init,
630
    .encode2        = svq1_encode_frame,
631
    .close          = svq1_encode_end,
632 633
    .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUV410P,
                                                     AV_PIX_FMT_NONE },
634
};