xref: /third_party/ffmpeg/libavcodec/zmbvenc.c (revision cabdff1a)
1/*
2 * Zip Motion Blocks Video (ZMBV) encoder
3 * Copyright (c) 2006 Konstantin Shishkov
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 * Zip Motion Blocks Video encoder
25 */
26
27#include <stdio.h>
28#include <stdlib.h>
29
30#include "libavutil/common.h"
31#include "libavutil/intreadwrite.h"
32#include "avcodec.h"
33#include "codec_internal.h"
34#include "encode.h"
35#include "zlib_wrapper.h"
36
37#include <zlib.h>
38
39/* Frame header flags */
40#define ZMBV_KEYFRAME 1
41#define ZMBV_DELTAPAL 2
42
43/* Motion block width/height (maximum allowed value is 255)
44 * Note: histogram datatype in block_cmp() must be big enough to hold values
45 * up to (4 * ZMBV_BLOCK * ZMBV_BLOCK)
46 */
47#define ZMBV_BLOCK 16
48
49/* Keyframe header format values */
50enum ZmbvFormat {
51    ZMBV_FMT_NONE  = 0,
52    ZMBV_FMT_1BPP  = 1,
53    ZMBV_FMT_2BPP  = 2,
54    ZMBV_FMT_4BPP  = 3,
55    ZMBV_FMT_8BPP  = 4,
56    ZMBV_FMT_15BPP = 5,
57    ZMBV_FMT_16BPP = 6,
58    ZMBV_FMT_24BPP = 7,
59    ZMBV_FMT_32BPP = 8
60};
61
62/**
63 * Encoder context
64 */
65typedef struct ZmbvEncContext {
66    AVCodecContext *avctx;
67
68    int lrange, urange;
69    uint8_t *comp_buf, *work_buf;
70    uint8_t pal[768];
71    uint32_t pal2[256]; //for quick comparisons
72    uint8_t *prev, *prev_buf;
73    int pstride;
74    int comp_size;
75    int keyint, curfrm;
76    int bypp;
77    enum ZmbvFormat fmt;
78    FFZStream zstream;
79
80    int score_tab[ZMBV_BLOCK * ZMBV_BLOCK * 4 + 1];
81} ZmbvEncContext;
82
83
84/** Block comparing function
85 * XXX should be optimized and moved to DSPContext
86 */
87static inline int block_cmp(ZmbvEncContext *c, uint8_t *src, int stride,
88                            uint8_t *src2, int stride2, int bw, int bh,
89                            int *xored)
90{
91    int sum = 0;
92    int i, j;
93    uint16_t histogram[256] = {0};
94    int bw_bytes = bw * c->bypp;
95
96    /* Build frequency histogram of byte values for src[] ^ src2[] */
97    for(j = 0; j < bh; j++){
98        for(i = 0; i < bw_bytes; i++){
99            int t = src[i] ^ src2[i];
100            histogram[t]++;
101        }
102        src += stride;
103        src2 += stride2;
104    }
105
106    /* If not all the xored values were 0, then the blocks are different */
107    *xored = (histogram[0] < bw_bytes * bh);
108
109    /* Exit early if blocks are equal */
110    if (!*xored) return 0;
111
112    /* Sum the entropy of all values */
113    for(i = 0; i < 256; i++)
114        sum += c->score_tab[histogram[i]];
115
116    return sum;
117}
118
119/** Motion estimation function
120 * TODO make better ME decisions
121 */
122static int zmbv_me(ZmbvEncContext *c, uint8_t *src, int sstride, uint8_t *prev,
123                   int pstride, int x, int y, int *mx, int *my, int *xored)
124{
125    int dx, dy, txored, tv, bv, bw, bh;
126    int mx0, my0;
127
128    mx0 = *mx;
129    my0 = *my;
130    bw = FFMIN(ZMBV_BLOCK, c->avctx->width - x);
131    bh = FFMIN(ZMBV_BLOCK, c->avctx->height - y);
132
133    /* Try (0,0) */
134    bv = block_cmp(c, src, sstride, prev, pstride, bw, bh, xored);
135    *mx = *my = 0;
136    if(!bv) return 0;
137
138    /* Try previous block's MV (if not 0,0) */
139    if (mx0 || my0){
140        tv = block_cmp(c, src, sstride, prev + mx0 * c->bypp + my0 * pstride, pstride, bw, bh, &txored);
141        if(tv < bv){
142            bv = tv;
143            *mx = mx0;
144            *my = my0;
145            *xored = txored;
146            if(!bv) return 0;
147        }
148    }
149
150    /* Try other MVs from top-to-bottom, left-to-right */
151    for(dy = -c->lrange; dy <= c->urange; dy++){
152        for(dx = -c->lrange; dx <= c->urange; dx++){
153            if(!dx && !dy) continue; // we already tested this block
154            if(dx == mx0 && dy == my0) continue; // this one too
155            tv = block_cmp(c, src, sstride, prev + dx * c->bypp + dy * pstride, pstride, bw, bh, &txored);
156            if(tv < bv){
157                 bv = tv;
158                 *mx = dx;
159                 *my = dy;
160                 *xored = txored;
161                 if(!bv) return 0;
162             }
163         }
164    }
165    return bv;
166}
167
168static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
169                        const AVFrame *pict, int *got_packet)
170{
171    ZmbvEncContext * const c = avctx->priv_data;
172    z_stream  *const zstream = &c->zstream.zstream;
173    const AVFrame * const p = pict;
174    uint8_t *src, *prev, *buf;
175    uint32_t *palptr;
176    int keyframe, chpal;
177    int fl;
178    int work_size = 0, pkt_size;
179    int bw, bh;
180    int i, j, ret;
181
182    keyframe = !c->curfrm;
183    c->curfrm++;
184    if(c->curfrm == c->keyint)
185        c->curfrm = 0;
186
187    palptr = (avctx->pix_fmt == AV_PIX_FMT_PAL8) ? (uint32_t *)p->data[1] : NULL;
188    chpal = !keyframe && palptr && memcmp(palptr, c->pal2, 1024);
189
190    src = p->data[0];
191    prev = c->prev;
192    if(chpal){
193        uint8_t tpal[3];
194        for(i = 0; i < 256; i++){
195            AV_WB24(tpal, palptr[i]);
196            c->work_buf[work_size++] = tpal[0] ^ c->pal[i * 3 + 0];
197            c->work_buf[work_size++] = tpal[1] ^ c->pal[i * 3 + 1];
198            c->work_buf[work_size++] = tpal[2] ^ c->pal[i * 3 + 2];
199            c->pal[i * 3 + 0] = tpal[0];
200            c->pal[i * 3 + 1] = tpal[1];
201            c->pal[i * 3 + 2] = tpal[2];
202        }
203        memcpy(c->pal2, palptr, 1024);
204    }
205    if(keyframe){
206        if (palptr){
207            for(i = 0; i < 256; i++){
208                AV_WB24(c->pal+(i*3), palptr[i]);
209            }
210            memcpy(c->work_buf, c->pal, 768);
211            memcpy(c->pal2, palptr, 1024);
212            work_size = 768;
213        }
214        for(i = 0; i < avctx->height; i++){
215            memcpy(c->work_buf + work_size, src, avctx->width * c->bypp);
216            src += p->linesize[0];
217            work_size += avctx->width * c->bypp;
218        }
219    }else{
220        int x, y, bh2, bw2, xored;
221        uint8_t *tsrc, *tprev;
222        uint8_t *mv;
223        int mx = 0, my = 0;
224
225        bw = (avctx->width + ZMBV_BLOCK - 1) / ZMBV_BLOCK;
226        bh = (avctx->height + ZMBV_BLOCK - 1) / ZMBV_BLOCK;
227        mv = c->work_buf + work_size;
228        memset(c->work_buf + work_size, 0, (bw * bh * 2 + 3) & ~3);
229        work_size += (bw * bh * 2 + 3) & ~3;
230        /* for now just XOR'ing */
231        for(y = 0; y < avctx->height; y += ZMBV_BLOCK) {
232            bh2 = FFMIN(avctx->height - y, ZMBV_BLOCK);
233            for(x = 0; x < avctx->width; x += ZMBV_BLOCK, mv += 2) {
234                bw2 = FFMIN(avctx->width - x, ZMBV_BLOCK);
235
236                tsrc = src + x * c->bypp;
237                tprev = prev + x * c->bypp;
238
239                zmbv_me(c, tsrc, p->linesize[0], tprev, c->pstride, x, y, &mx, &my, &xored);
240                mv[0] = (mx * 2) | !!xored;
241                mv[1] = my * 2;
242                tprev += mx * c->bypp + my * c->pstride;
243                if(xored){
244                    for(j = 0; j < bh2; j++){
245                        for(i = 0; i < bw2 * c->bypp; i++)
246                            c->work_buf[work_size++] = tsrc[i] ^ tprev[i];
247                        tsrc += p->linesize[0];
248                        tprev += c->pstride;
249                    }
250                }
251            }
252            src += p->linesize[0] * ZMBV_BLOCK;
253            prev += c->pstride * ZMBV_BLOCK;
254        }
255    }
256    /* save the previous frame */
257    src = p->data[0];
258    prev = c->prev;
259    for(i = 0; i < avctx->height; i++){
260        memcpy(prev, src, avctx->width * c->bypp);
261        prev += c->pstride;
262        src += p->linesize[0];
263    }
264
265    if (keyframe)
266        deflateReset(zstream);
267
268    zstream->next_in   = c->work_buf;
269    zstream->avail_in  = work_size;
270    zstream->total_in  = 0;
271
272    zstream->next_out  = c->comp_buf;
273    zstream->avail_out = c->comp_size;
274    zstream->total_out = 0;
275    if (deflate(zstream, Z_SYNC_FLUSH) != Z_OK) {
276        av_log(avctx, AV_LOG_ERROR, "Error compressing data\n");
277        return -1;
278    }
279
280    pkt_size = zstream->total_out + 1 + 6 * keyframe;
281    if ((ret = ff_get_encode_buffer(avctx, pkt, pkt_size, 0)) < 0)
282        return ret;
283    buf = pkt->data;
284
285    fl = (keyframe ? ZMBV_KEYFRAME : 0) | (chpal ? ZMBV_DELTAPAL : 0);
286    *buf++ = fl;
287    if (keyframe) {
288        *buf++ = 0; // hi ver
289        *buf++ = 1; // lo ver
290        *buf++ = 1; // comp
291        *buf++ = c->fmt; // format
292        *buf++ = ZMBV_BLOCK; // block width
293        *buf++ = ZMBV_BLOCK; // block height
294        pkt->flags |= AV_PKT_FLAG_KEY;
295    }
296    memcpy(buf, c->comp_buf, zstream->total_out);
297
298    *got_packet = 1;
299
300    return 0;
301}
302
303static av_cold int encode_end(AVCodecContext *avctx)
304{
305    ZmbvEncContext * const c = avctx->priv_data;
306
307    av_freep(&c->comp_buf);
308    av_freep(&c->work_buf);
309
310    av_freep(&c->prev_buf);
311    ff_deflate_end(&c->zstream);
312
313    return 0;
314}
315
316/**
317 * Init zmbv encoder
318 */
319static av_cold int encode_init(AVCodecContext *avctx)
320{
321    ZmbvEncContext * const c = avctx->priv_data;
322    int i;
323    int lvl = 9;
324    int prev_size, prev_offset;
325
326    switch (avctx->pix_fmt) {
327    case AV_PIX_FMT_PAL8:
328        c->fmt = ZMBV_FMT_8BPP;
329        c->bypp = 1;
330        break;
331    case AV_PIX_FMT_RGB555LE:
332        c->fmt = ZMBV_FMT_15BPP;
333        c->bypp = 2;
334        break;
335    case AV_PIX_FMT_RGB565LE:
336        c->fmt = ZMBV_FMT_16BPP;
337        c->bypp = 2;
338        break;
339#ifdef ZMBV_ENABLE_24BPP
340    case AV_PIX_FMT_BGR24:
341        c->fmt = ZMBV_FMT_24BPP;
342        c->bypp = 3;
343        break;
344#endif //ZMBV_ENABLE_24BPP
345    case AV_PIX_FMT_BGR0:
346        c->fmt = ZMBV_FMT_32BPP;
347        c->bypp = 4;
348        break;
349    }
350
351    /* Entropy-based score tables for comparing blocks.
352     * Suitable for blocks up to (ZMBV_BLOCK * ZMBV_BLOCK) bytes.
353     * Scores are nonnegative, lower is better.
354     */
355    for(i = 1; i <= ZMBV_BLOCK * ZMBV_BLOCK * c->bypp; i++)
356        c->score_tab[i] = -i * log2(i / (double)(ZMBV_BLOCK * ZMBV_BLOCK * c->bypp)) * 256;
357
358    c->avctx = avctx;
359
360    c->curfrm = 0;
361    c->keyint = avctx->keyint_min;
362
363    /* Motion estimation range: maximum distance is -64..63 */
364    c->lrange = c->urange = 8;
365    if(avctx->me_range > 0){
366        c->lrange = FFMIN(avctx->me_range, 64);
367        c->urange = FFMIN(avctx->me_range, 63);
368    }
369
370    if(avctx->compression_level >= 0)
371        lvl = avctx->compression_level;
372    if(lvl < 0 || lvl > 9){
373        av_log(avctx, AV_LOG_ERROR, "Compression level should be 0-9, not %i\n", lvl);
374        return AVERROR(EINVAL);
375    }
376
377    c->comp_size = avctx->width * c->bypp * avctx->height + 1024 +
378        ((avctx->width + ZMBV_BLOCK - 1) / ZMBV_BLOCK) * ((avctx->height + ZMBV_BLOCK - 1) / ZMBV_BLOCK) * 2 + 4;
379    if (!(c->work_buf = av_malloc(c->comp_size))) {
380        av_log(avctx, AV_LOG_ERROR, "Can't allocate work buffer.\n");
381        return AVERROR(ENOMEM);
382    }
383    /* Conservative upper bound taken from zlib v1.2.1 source via lcl.c */
384    c->comp_size = c->comp_size + ((c->comp_size + 7) >> 3) +
385                           ((c->comp_size + 63) >> 6) + 11;
386
387    /* Allocate compression buffer */
388    if (!(c->comp_buf = av_malloc(c->comp_size))) {
389        av_log(avctx, AV_LOG_ERROR, "Can't allocate compression buffer.\n");
390        return AVERROR(ENOMEM);
391    }
392
393    /* Allocate prev buffer - pad around the image to allow out-of-edge ME:
394     * - The image should be padded with `lrange` rows before and `urange` rows
395     *   after.
396     * - The stride should be padded with `lrange` pixels, then rounded up to a
397     *   multiple of 16 bytes.
398     * - The first row should also be padded with `lrange` pixels before, then
399     *   aligned up to a multiple of 16 bytes.
400     */
401    c->pstride = FFALIGN((avctx->width + c->lrange) * c->bypp, 16);
402    prev_size = FFALIGN(c->lrange * c->bypp, 16) + c->pstride * (c->lrange + avctx->height + c->urange);
403    prev_offset = FFALIGN(c->lrange * c->bypp, 16) + c->pstride * c->lrange;
404    if (!(c->prev_buf = av_mallocz(prev_size))) {
405        av_log(avctx, AV_LOG_ERROR, "Can't allocate picture.\n");
406        return AVERROR(ENOMEM);
407    }
408    c->prev = c->prev_buf + prev_offset;
409
410    return ff_deflate_init(&c->zstream, lvl, avctx);
411}
412
413const FFCodec ff_zmbv_encoder = {
414    .p.name         = "zmbv",
415    .p.long_name    = NULL_IF_CONFIG_SMALL("Zip Motion Blocks Video"),
416    .p.type         = AVMEDIA_TYPE_VIDEO,
417    .p.id           = AV_CODEC_ID_ZMBV,
418    .p.capabilities = AV_CODEC_CAP_DR1,
419    .priv_data_size = sizeof(ZmbvEncContext),
420    .init           = encode_init,
421    FF_CODEC_ENCODE_CB(encode_frame),
422    .close          = encode_end,
423    .p.pix_fmts     = (const enum AVPixelFormat[]) { AV_PIX_FMT_PAL8,
424                                                     AV_PIX_FMT_RGB555LE,
425                                                     AV_PIX_FMT_RGB565LE,
426#ifdef ZMBV_ENABLE_24BPP
427                                                     AV_PIX_FMT_BGR24,
428#endif //ZMBV_ENABLE_24BPP
429                                                     AV_PIX_FMT_BGR0,
430                                                     AV_PIX_FMT_NONE },
431    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
432};
433