2 * Flash Screen Video encoder
3 * Copyright (C) 2004 Alex Beregszaszi
4 * Copyright (C) 2006 Benjamin Larsson
6 * This file is part of Libav.
8 * Libav 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.
13 * Libav 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.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 /* Encoding development sponsored by http://fh-campuswien.ac.at */
27 * Flash Screen Video encoder
28 * @author Alex Beregszaszi
29 * @author Benjamin Larsson
32 /* Bitstream description
33 * The picture is divided into blocks that are zlib-compressed.
35 * The decoder is fed complete frames, the frameheader contains:
36 * 4 bits of block width
37 * 12 bits of frame width
38 * 4 bits of block height
39 * 12 bits of frame height
41 * Directly after the header are the compressed blocks. The blocks
42 * have their compressed size represented with 16 bits in the beginning.
43 * If the size = 0 then the block is unchanged from the previous frame.
44 * All blocks are decompressed until the buffer is consumed.
46 * Encoding ideas: A basic encoder would just use a fixed block size.
47 * Block sizes can be multiples 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.
52 * Don't reencode the frame in brute force mode if the frame is a dupe.
53 * Speed up. Make the difference check faster.
62 #include "bytestream.h"
65 typedef struct FlashSVContext {
66 AVCodecContext *avctx;
67 uint8_t *previous_frame;
69 int image_width, image_height;
70 int block_width, block_height;
78 static int copy_region_enc(uint8_t *sptr, uint8_t *dptr, int dx, int dy,
79 int h, int w, int stride, uint8_t *pfptr)
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];
100 static av_cold int flashsv_encode_init(AVCodecContext *avctx)
102 FlashSVContext *s = avctx->priv_data;
106 if (avctx->width > 4095 || avctx->height > 4095) {
107 av_log(avctx, AV_LOG_ERROR,
108 "Input dimensions too large, input must be max 4096x4096 !\n");
109 return AVERROR_INVALIDDATA;
112 // Needed if zlib unused or init aborted before deflateInit
113 memset(&s->zstream, 0, sizeof(z_stream));
115 s->last_key_frame = 0;
117 s->image_width = avctx->width;
118 s->image_height = avctx->height;
120 s->tmpblock = av_mallocz(3 * 256 * 256);
121 s->encbuffer = av_mallocz(s->image_width * s->image_height * 3);
123 if (!s->tmpblock || !s->encbuffer) {
124 av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n");
125 return AVERROR(ENOMEM);
132 static int encode_bitstream(FlashSVContext *s, AVFrame *p, uint8_t *buf,
133 int buf_size, int block_width, int block_height,
134 uint8_t *previous_frame, int *I_frame)
138 int h_blocks, v_blocks, h_part, v_part, i, j;
142 init_put_bits(&pb, buf, buf_size * 8);
144 put_bits(&pb, 4, block_width / 16 - 1);
145 put_bits(&pb, 12, s->image_width);
146 put_bits(&pb, 4, block_height / 16 - 1);
147 put_bits(&pb, 12, s->image_height);
151 h_blocks = s->image_width / block_width;
152 h_part = s->image_width % block_width;
153 v_blocks = s->image_height / block_height;
154 v_part = s->image_height % block_height;
156 /* loop over all block columns */
157 for (j = 0; j < v_blocks + (v_part ? 1 : 0); j++) {
159 int hp = j * block_height; // horizontal position in frame
160 int hs = (j < v_blocks) ? block_height : v_part; // size of block
162 /* loop over all block rows */
163 for (i = 0; i < h_blocks + (h_part ? 1 : 0); i++) {
164 int wp = i * block_width; // vertical position in frame
165 int ws = (i < h_blocks) ? block_width : h_part; // size of block
167 uint8_t *ptr = buf + buf_pos;
169 /* copy the block to the temp buffer before compression
170 * (if it differs from the previous frame's block) */
171 res = copy_region_enc(p->data[0], s->tmpblock,
172 s->image_height - (hp + hs + 1),
173 wp, hs, ws, p->linesize[0], previous_frame);
175 if (res || *I_frame) {
176 unsigned long zsize = 3 * block_width * block_height;
177 ret = compress2(ptr + 2, &zsize, s->tmpblock, 3 * ws * hs, 9);
179 //ret = deflateReset(&s->zstream);
181 av_log(s->avctx, AV_LOG_ERROR,
182 "error while compressing block %dx%d\n", i, j);
184 bytestream_put_be16(&ptr, (unsigned int) zsize);
185 buf_pos += zsize + 2;
186 av_dlog(avctx, "buf_pos = %d\n", buf_pos);
189 bytestream_put_be16(&ptr, 0);
204 static int flashsv_encode_frame(AVCodecContext *avctx, uint8_t *buf,
205 int buf_size, void *data)
207 FlashSVContext * const s = avctx->priv_data;
208 AVFrame *pict = data;
209 AVFrame * const p = &s->frame;
213 int opt_w = 4, opt_h = 4;
217 /* First frame needs to be a keyframe */
218 if (avctx->frame_number == 0) {
219 s->previous_frame = av_mallocz(FFABS(p->linesize[0]) * s->image_height);
220 if (!s->previous_frame) {
221 av_log(avctx, AV_LOG_ERROR, "Memory allocation failed.\n");
222 return AVERROR(ENOMEM);
227 if (p->linesize[0] < 0)
228 pfptr = s->previous_frame - (s->image_height - 1) * p->linesize[0];
230 pfptr = s->previous_frame;
232 /* Check the placement of keyframes */
233 if (avctx->gop_size > 0 &&
234 avctx->frame_number >= s->last_key_frame + avctx->gop_size) {
238 if (buf_size < s->image_width * s->image_height * 3) {
239 //Conservative upper bound check for compressed data
240 av_log(avctx, AV_LOG_ERROR, "buf_size %d < %d\n",
241 buf_size, s->image_width * s->image_height * 3);
245 res = encode_bitstream(s, p, buf, buf_size, opt_w * 16, opt_h * 16,
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]);
252 memcpy(s->previous_frame,
253 p->data[0] + p->linesize[0] * (s->image_height - 1),
254 s->image_height * FFABS(p->linesize[0]));
256 //mark the frame type so the muxer can mux it correctly
258 p->pict_type = AV_PICTURE_TYPE_I;
260 s->last_key_frame = avctx->frame_number;
261 av_dlog(avctx, "Inserting keyframe at frame %d\n", avctx->frame_number);
263 p->pict_type = AV_PICTURE_TYPE_P;
267 avctx->coded_frame = p;
272 static av_cold int flashsv_encode_end(AVCodecContext *avctx)
274 FlashSVContext *s = avctx->priv_data;
276 deflateEnd(&s->zstream);
278 av_free(s->encbuffer);
279 av_free(s->previous_frame);
280 av_free(s->tmpblock);
285 AVCodec ff_flashsv_encoder = {
287 .type = AVMEDIA_TYPE_VIDEO,
288 .id = CODEC_ID_FLASHSV,
289 .priv_data_size = sizeof(FlashSVContext),
290 .init = flashsv_encode_init,
291 .encode = flashsv_encode_frame,
292 .close = flashsv_encode_end,
293 .pix_fmts = (const enum PixelFormat[]){PIX_FMT_BGR24, PIX_FMT_NONE},
294 .long_name = NULL_IF_CONFIG_SMALL("Flash Screen Video"),