1/*
2 * Flash Screen Video encoder
3 * Copyright (C) 2004 Alex Beregszaszi
4 * Copyright (C) 2006 Benjamin Larsson
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23/* Encoding development sponsored by http://fh-campuswien.ac.at */
24
25/**
26 * @file
27 * Flash Screen Video encoder
28 * @author Alex Beregszaszi
29 * @author Benjamin Larsson
30 */
31
32/* Bitstream description
33 * The picture is divided into blocks that are zlib-compressed.
34 *
35 * The decoder is fed complete frames, the frameheader contains:
36 * 4bits of block width
37 * 12bits of frame width
38 * 4bits of block height
39 * 12bits of frame height
40 *
41 * Directly after the header are the compressed blocks. The blocks
42 * have their compressed size represented with 16bits in the beginig.
43 * If the size = 0 then the block is unchanged from the previous frame.
44 * All blocks are decompressed until the buffer is consumed.
45 *
46 * Encoding ideas, a basic encoder would just use a fixed block size.
47 * Block sizes can be multipels of 16, from 16 to 256. The blocks don't
48 * have to be quadratic. A brute force search with a set of different
49 * block sizes should give a better result than to just use a fixed size.
50 */
51
52/* TODO:
53 * Don't reencode the frame in brute force mode if the frame is a dupe. Speed up.
54 * Make the difference check faster.
55 */
56
57#include <stdio.h>
58#include <stdlib.h>
59#include <zlib.h>
60
61#include "avcodec.h"
62#include "put_bits.h"
63#include "bytestream.h"
64
65
66typedef struct FlashSVContext {
67    AVCodecContext *avctx;
68    uint8_t *previous_frame;
69    AVFrame frame;
70    int image_width, image_height;
71    int block_width, block_height;
72    uint8_t* tmpblock;
73    uint8_t* encbuffer;
74    int block_size;
75    z_stream zstream;
76    int last_key_frame;
77} FlashSVContext;
78
79static int copy_region_enc(uint8_t *sptr, uint8_t *dptr,
80        int dx, int dy, int h, int w, int stride, uint8_t *pfptr) {
81    int i,j;
82    uint8_t *nsptr;
83    uint8_t *npfptr;
84    int diff = 0;
85
86    for (i = dx+h; i > dx; i--) {
87        nsptr = sptr+(i*stride)+dy*3;
88        npfptr = pfptr+(i*stride)+dy*3;
89        for (j=0 ; j<w*3 ; j++) {
90            diff |=npfptr[j]^nsptr[j];
91            dptr[j] = nsptr[j];
92        }
93        dptr += w*3;
94    }
95    if (diff)
96        return 1;
97    return 0;
98}
99
100static av_cold int flashsv_encode_init(AVCodecContext *avctx)
101{
102    FlashSVContext *s = avctx->priv_data;
103
104    s->avctx = avctx;
105
106    if ((avctx->width > 4095) || (avctx->height > 4095)) {
107        av_log(avctx, AV_LOG_ERROR, "Input dimensions too large, input must be max 4096x4096 !\n");
108        return -1;
109    }
110
111    // Needed if zlib unused or init aborted before deflateInit
112    memset(&(s->zstream), 0, sizeof(z_stream));
113
114    s->last_key_frame=0;
115
116    s->image_width = avctx->width;
117    s->image_height = avctx->height;
118
119    s->tmpblock = av_mallocz(3*256*256);
120    s->encbuffer = av_mallocz(s->image_width*s->image_height*3);
121
122    if (!s->tmpblock || !s->encbuffer) {
123        av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n");
124        return -1;
125    }
126
127    return 0;
128}
129
130
131static int encode_bitstream(FlashSVContext *s, AVFrame *p, uint8_t *buf, int buf_size,
132     int block_width, int block_height, uint8_t *previous_frame, int* I_frame) {
133
134    PutBitContext pb;
135    int h_blocks, v_blocks, h_part, v_part, i, j;
136    int buf_pos, res;
137    int pred_blocks = 0;
138
139    init_put_bits(&pb, buf, buf_size*8);
140
141    put_bits(&pb, 4, (block_width/16)-1);
142    put_bits(&pb, 12, s->image_width);
143    put_bits(&pb, 4, (block_height/16)-1);
144    put_bits(&pb, 12, s->image_height);
145    flush_put_bits(&pb);
146    buf_pos=4;
147
148    h_blocks = s->image_width / block_width;
149    h_part = s->image_width % block_width;
150    v_blocks = s->image_height / block_height;
151    v_part = s->image_height % block_height;
152
153    /* loop over all block columns */
154    for (j = 0; j < v_blocks + (v_part?1:0); j++)
155    {
156
157        int hp = j*block_height; // horiz position in frame
158        int hs = (j<v_blocks)?block_height:v_part; // size of block
159
160        /* loop over all block rows */
161        for (i = 0; i < h_blocks + (h_part?1:0); i++)
162        {
163            int wp = i*block_width; // vert position in frame
164            int ws = (i<h_blocks)?block_width:h_part; // size of block
165            int ret=Z_OK;
166            uint8_t *ptr;
167
168            ptr = buf+buf_pos;
169
170            //copy the block to the temp buffer before compression (if it differs from the previous frame's block)
171            res = copy_region_enc(p->data[0], s->tmpblock, s->image_height-(hp+hs+1), wp, hs, ws, p->linesize[0], previous_frame);
172
173            if (res || *I_frame) {
174                unsigned long zsize;
175                zsize = 3*block_width*block_height;
176                ret = compress2(ptr+2, &zsize, s->tmpblock, 3*ws*hs, 9);
177
178
179                //ret = deflateReset(&(s->zstream));
180                if (ret != Z_OK)
181                    av_log(s->avctx, AV_LOG_ERROR, "error while compressing block %dx%d\n", i, j);
182
183                bytestream_put_be16(&ptr,(unsigned int)zsize);
184                buf_pos += zsize+2;
185                //av_log(avctx, AV_LOG_ERROR, "buf_pos = %d\n", buf_pos);
186            } else {
187                pred_blocks++;
188                bytestream_put_be16(&ptr,0);
189                buf_pos += 2;
190            }
191        }
192    }
193
194    if (pred_blocks)
195        *I_frame = 0;
196    else
197        *I_frame = 1;
198
199    return buf_pos;
200}
201
202
203static int flashsv_encode_frame(AVCodecContext *avctx, uint8_t *buf, int buf_size, void *data)
204{
205    FlashSVContext * const s = avctx->priv_data;
206    AVFrame *pict = data;
207    AVFrame * const p = &s->frame;
208    uint8_t *pfptr;
209    int res;
210    int I_frame = 0;
211    int opt_w, opt_h;
212
213    *p = *pict;
214
215    /* First frame needs to be a keyframe */
216    if (avctx->frame_number == 0) {
217        s->previous_frame = av_mallocz(FFABS(p->linesize[0])*s->image_height);
218        if (!s->previous_frame) {
219            av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n");
220            return -1;
221        }
222        I_frame = 1;
223    }
224
225    if (p->linesize[0] < 0)
226        pfptr = s->previous_frame - ((s->image_height-1) * p->linesize[0]);
227    else
228        pfptr = s->previous_frame;
229
230    /* Check the placement of keyframes */
231    if (avctx->gop_size > 0) {
232        if (avctx->frame_number >= s->last_key_frame + avctx->gop_size) {
233            I_frame = 1;
234        }
235    }
236
237    opt_w=4;
238    opt_h=4;
239
240    if (buf_size < s->image_width*s->image_height*3) {
241        //Conservative upper bound check for compressed data
242        av_log(avctx, AV_LOG_ERROR, "buf_size %d <  %d\n", buf_size, s->image_width*s->image_height*3);
243        return -1;
244    }
245
246    res = encode_bitstream(s, p, buf, buf_size, opt_w*16, opt_h*16, pfptr, &I_frame);
247
248    //save the current frame
249    if(p->linesize[0] > 0)
250        memcpy(s->previous_frame, p->data[0], s->image_height*p->linesize[0]);
251    else
252        memcpy(s->previous_frame, p->data[0] + p->linesize[0] * (s->image_height-1), s->image_height*FFABS(p->linesize[0]));
253
254    //mark the frame type so the muxer can mux it correctly
255    if (I_frame) {
256        p->pict_type = FF_I_TYPE;
257        p->key_frame = 1;
258        s->last_key_frame = avctx->frame_number;
259        av_log(avctx, AV_LOG_DEBUG, "Inserting key frame at frame %d\n",avctx->frame_number);
260    } else {
261        p->pict_type = FF_P_TYPE;
262        p->key_frame = 0;
263    }
264
265    avctx->coded_frame = p;
266
267    return res;
268}
269
270static av_cold int flashsv_encode_end(AVCodecContext *avctx)
271{
272    FlashSVContext *s = avctx->priv_data;
273
274    deflateEnd(&(s->zstream));
275
276    av_free(s->encbuffer);
277    av_free(s->previous_frame);
278    av_free(s->tmpblock);
279
280    return 0;
281}
282
283AVCodec flashsv_encoder = {
284    "flashsv",
285    AVMEDIA_TYPE_VIDEO,
286    CODEC_ID_FLASHSV,
287    sizeof(FlashSVContext),
288    flashsv_encode_init,
289    flashsv_encode_frame,
290    flashsv_encode_end,
291    .pix_fmts = (const enum PixelFormat[]){PIX_FMT_BGR24, PIX_FMT_NONE},
292    .long_name = NULL_IF_CONFIG_SMALL("Flash Screen Video"),
293};
294
295