1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-02 03:06:28 +02:00
FFmpeg/libavcodec/mjpegenc.c
Jerry Jiang 884506dfe2 Implement optimal huffman encoding for (M)JPEG.
> seems to break
> make fate-vsynth1-mjpeg-444

Fixed.
2017-02-08 13:59:53 +00:00

429 lines
14 KiB
C

/*
* MJPEG encoder
* Copyright (c) 2000, 2001 Fabrice Bellard
* Copyright (c) 2003 Alex Beregszaszi
* Copyright (c) 2003-2004 Michael Niedermayer
*
* Support for external huffman table, various fixes (AVID workaround),
* aspecting, new decode_frame mechanism and apple mjpeg-b support
* by Alex Beregszaszi
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* 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.
*
* FFmpeg is distributed in the hope that it will be useful,
* 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
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* MJPEG encoder.
*/
#include "libavutil/pixdesc.h"
#include "avcodec.h"
#include "jpegtables.h"
#include "mjpegenc_common.h"
#include "mpegvideo.h"
#include "mjpeg.h"
#include "mjpegenc.h"
// Don't know, but let's guess 16 bits per code
#define MJPEG_HUFFMAN_EST_BITS_PER_CODE 16
av_cold int ff_mjpeg_encode_init(MpegEncContext *s)
{
MJpegContext *m;
av_assert0(s->slice_context_count == 1);
if (s->width > 65500 || s->height > 65500) {
av_log(s, AV_LOG_ERROR, "JPEG does not support resolutions above 65500x65500\n");
return AVERROR(EINVAL);
}
m = av_malloc(sizeof(MJpegContext));
if (!m)
return AVERROR(ENOMEM);
s->min_qcoeff=-1023;
s->max_qcoeff= 1023;
// Build default Huffman tables.
// These may be overwritten later with more optimal Huffman tables, but
// they are needed at least right now for some processes like trellis.
ff_mjpeg_build_huffman_codes(m->huff_size_dc_luminance,
m->huff_code_dc_luminance,
avpriv_mjpeg_bits_dc_luminance,
avpriv_mjpeg_val_dc);
ff_mjpeg_build_huffman_codes(m->huff_size_dc_chrominance,
m->huff_code_dc_chrominance,
avpriv_mjpeg_bits_dc_chrominance,
avpriv_mjpeg_val_dc);
ff_mjpeg_build_huffman_codes(m->huff_size_ac_luminance,
m->huff_code_ac_luminance,
avpriv_mjpeg_bits_ac_luminance,
avpriv_mjpeg_val_ac_luminance);
ff_mjpeg_build_huffman_codes(m->huff_size_ac_chrominance,
m->huff_code_ac_chrominance,
avpriv_mjpeg_bits_ac_chrominance,
avpriv_mjpeg_val_ac_chrominance);
init_uni_ac_vlc(m->huff_size_ac_luminance, m->uni_ac_vlc_len);
init_uni_ac_vlc(m->huff_size_ac_chrominance, m->uni_chroma_ac_vlc_len);
s->intra_ac_vlc_length =
s->intra_ac_vlc_last_length = m->uni_ac_vlc_len;
s->intra_chroma_ac_vlc_length =
s->intra_chroma_ac_vlc_last_length = m->uni_chroma_ac_vlc_len;
// Buffers start out empty.
m->huff_buffer = NULL;
m->huff_ncode = 0;
m->huff_capacity = 0;
m->error = 0;
s->mjpeg_ctx = m;
return 0;
}
av_cold void ff_mjpeg_encode_close(MpegEncContext *s)
{
av_freep(&s->mjpeg_ctx->huff_buffer);
av_freep(&s->mjpeg_ctx);
}
/**
* Encodes and outputs the entire frame in the JPEG format.
*
* @param s The MpegEncContext.
*/
void ff_mjpeg_encode_picture_frame(MpegEncContext *s)
{
int i, nbits, code, table_id;
MJpegContext *m = s->mjpeg_ctx;
uint8_t *huff_size[4] = {m->huff_size_dc_luminance,
m->huff_size_dc_chrominance,
m->huff_size_ac_luminance,
m->huff_size_ac_chrominance};
uint16_t *huff_code[4] = {m->huff_code_dc_luminance,
m->huff_code_dc_chrominance,
m->huff_code_ac_luminance,
m->huff_code_ac_chrominance};
size_t total_bits = 0;
size_t bytes_needed;
// Estimate the total size first
for (i = 0; i < m->huff_ncode; i++) {
table_id = m->huff_buffer[i].table_id;
code = m->huff_buffer[i].code;
nbits = code & 0xf;
total_bits += huff_size[table_id][code] + nbits;
}
bytes_needed = (total_bits + 7) / 8;
ff_mpv_reallocate_putbitbuffer(s, bytes_needed, bytes_needed);
for (i = 0; i < m->huff_ncode; i++) {
table_id = m->huff_buffer[i].table_id;
code = m->huff_buffer[i].code;
nbits = code & 0xf;
put_bits(&s->pb, huff_size[table_id][code], huff_code[table_id][code]);
if (nbits != 0) {
put_sbits(&s->pb, nbits, m->huff_buffer[i].mant);
}
}
m->huff_ncode = 0;
}
/**
* Add code and table_id to the JPEG buffer.
*
* @param s The MJpegContext which contains the JPEG buffer.
* @param table_id Which Huffman table the code belongs to.
* @param code The encoded exponent of the coefficients and the run-bits.
*/
static inline void ff_mjpeg_encode_code(MJpegContext *s, uint8_t table_id, int code)
{
MJpegHuffmanCode *c = &s->huff_buffer[s->huff_ncode++];
av_assert0(s->huff_ncode < s->huff_capacity);
c->table_id = table_id;
c->code = code;
}
/**
* Add the coefficient's data to the JPEG buffer.
*
* @param s The MJpegContext which contains the JPEG buffer.
* @param table_id Which Huffman table the code belongs to.
* @param val The coefficient.
* @param run The run-bits.
*/
static void ff_mjpeg_encode_coef(MJpegContext *s, uint8_t table_id, int val, int run)
{
int mant, code;
if (val == 0) {
av_assert0(run == 0);
ff_mjpeg_encode_code(s, table_id, 0);
} else {
mant = val;
if (val < 0) {
val = -val;
mant--;
}
code = (run << 4) | (av_log2_16bit(val) + 1);
s->huff_buffer[s->huff_ncode].mant = mant;
ff_mjpeg_encode_code(s, table_id, code);
}
}
/**
* Add the block's data into the JPEG buffer.
*
* @param s The MJpegEncContext that contains the JPEG buffer.
* @param block The block.
* @param n The block's index or number.
*/
static void encode_block(MpegEncContext *s, int16_t *block, int n)
{
int i, j, table_id;
int component, dc, last_index, val, run;
MJpegContext *m = s->mjpeg_ctx;
if (m->error) return;
av_assert0(m->huff_capacity >= m->huff_ncode + 64);
/* DC coef */
component = (n <= 3 ? 0 : (n&1) + 1);
table_id = (n <= 3 ? 0 : 1);
dc = block[0]; /* overflow is impossible */
val = dc - s->last_dc[component];
ff_mjpeg_encode_coef(m, table_id, val, 0);
s->last_dc[component] = dc;
/* AC coefs */
run = 0;
last_index = s->block_last_index[n];
table_id |= 2;
for(i=1;i<=last_index;i++) {
j = s->intra_scantable.permutated[i];
val = block[j];
if (val == 0) {
run++;
} else {
while (run >= 16) {
ff_mjpeg_encode_code(m, table_id, 0xf0);
run -= 16;
}
ff_mjpeg_encode_coef(m, table_id, val, run);
run = 0;
}
}
/* output EOB only if not already 64 values */
if (last_index < 63 || run != 0)
ff_mjpeg_encode_code(m, table_id, 0);
}
// Possibly reallocate the huffman code buffer, assuming blocks_per_mb.
// Set s->mjpeg_ctx->error on ENOMEM.
static void realloc_huffman(MpegEncContext *s, int blocks_per_mb)
{
MJpegContext *m = s->mjpeg_ctx;
size_t num_mbs, num_blocks, num_codes;
MJpegHuffmanCode *new_buf;
if (m->error) return;
// Make sure we have enough space to hold this frame.
num_mbs = s->mb_width * s->mb_height;
num_blocks = num_mbs * blocks_per_mb;
av_assert0(m->huff_ncode <=
(s->mb_y * s->mb_width + s->mb_x) * blocks_per_mb * 64);
num_codes = num_blocks * 64;
new_buf = av_fast_realloc(m->huff_buffer, &m->huff_capacity,
num_codes * sizeof(MJpegHuffmanCode));
if (!new_buf) {
m->error = AVERROR(ENOMEM);
} else {
m->huff_buffer = new_buf;
}
}
int ff_mjpeg_encode_mb(MpegEncContext *s, int16_t block[12][64])
{
int i, is_chroma_420;
// Number of bits used depends on future data.
// So, nothing that relies on encoding many times and taking the
// one with the fewest bits will work properly here.
if (s->i_tex_bits != MJPEG_HUFFMAN_EST_BITS_PER_CODE *
s->mjpeg_ctx->huff_ncode) {
av_log(NULL, AV_LOG_ERROR, "Unsupported encoding method\n");
return AVERROR(EINVAL);
}
if (s->chroma_format == CHROMA_444) {
realloc_huffman(s, 12);
encode_block(s, block[0], 0);
encode_block(s, block[2], 2);
encode_block(s, block[4], 4);
encode_block(s, block[8], 8);
encode_block(s, block[5], 5);
encode_block(s, block[9], 9);
if (16*s->mb_x+8 < s->width) {
encode_block(s, block[1], 1);
encode_block(s, block[3], 3);
encode_block(s, block[6], 6);
encode_block(s, block[10], 10);
encode_block(s, block[7], 7);
encode_block(s, block[11], 11);
}
} else {
is_chroma_420 = (s->chroma_format == CHROMA_420);
realloc_huffman(s, 5 + (is_chroma_420 ? 1 : 3));
for(i=0;i<5;i++) {
encode_block(s, block[i], i);
}
if (is_chroma_420) {
encode_block(s, block[5], 5);
} else {
encode_block(s, block[6], 6);
encode_block(s, block[5], 5);
encode_block(s, block[7], 7);
}
}
if (s->mjpeg_ctx->error)
return s->mjpeg_ctx->error;
s->i_tex_bits = MJPEG_HUFFMAN_EST_BITS_PER_CODE * s->mjpeg_ctx->huff_ncode;
return 0;
}
// maximum over s->mjpeg_vsample[i]
#define V_MAX 2
static int amv_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
const AVFrame *pic_arg, int *got_packet)
{
MpegEncContext *s = avctx->priv_data;
AVFrame *pic;
int i, ret;
int chroma_h_shift, chroma_v_shift;
av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
#if FF_API_EMU_EDGE
//CODEC_FLAG_EMU_EDGE have to be cleared
if(s->avctx->flags & CODEC_FLAG_EMU_EDGE)
return AVERROR(EINVAL);
#endif
if ((avctx->height & 15) && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) {
av_log(avctx, AV_LOG_ERROR,
"Heights which are not a multiple of 16 might fail with some decoders, "
"use vstrict=-1 / -strict -1 to use %d anyway.\n", avctx->height);
av_log(avctx, AV_LOG_WARNING, "If you have a device that plays AMV videos, please test if videos "
"with such heights work with it and report your findings to ffmpeg-devel@ffmpeg.org\n");
return AVERROR_EXPERIMENTAL;
}
pic = av_frame_clone(pic_arg);
if (!pic)
return AVERROR(ENOMEM);
//picture should be flipped upside-down
for(i=0; i < 3; i++) {
int vsample = i ? 2 >> chroma_v_shift : 2;
pic->data[i] += pic->linesize[i] * (vsample * s->height / V_MAX - 1);
pic->linesize[i] *= -1;
}
ret = ff_mpv_encode_picture(avctx, pkt, pic, got_packet);
av_frame_free(&pic);
return ret;
}
#define OFFSET(x) offsetof(MpegEncContext, x)
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
static const AVOption options[] = {
FF_MPV_COMMON_OPTS
{ "pred", "Prediction method", OFFSET(pred), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 3, VE, "pred" },
{ "left", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, INT_MIN, INT_MAX, VE, "pred" },
{ "plane", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, INT_MIN, INT_MAX, VE, "pred" },
{ "median", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = 3 }, INT_MIN, INT_MAX, VE, "pred" },
{ "huffman", "Huffman table strategy", OFFSET(huffman), AV_OPT_TYPE_INT, { .i64 = HUFFMAN_TABLE_DEFAULT }, 0, NB_HUFFMAN_TABLE_OPTION - 1, VE, "huffman" },
{ "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_DEFAULT }, INT_MIN, INT_MAX, VE, "huffman" },
{ "optimal", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = HUFFMAN_TABLE_OPTIMAL }, INT_MIN, INT_MAX, VE, "huffman" },
{ NULL},
};
#if CONFIG_MJPEG_ENCODER
static const AVClass mjpeg_class = {
.class_name = "mjpeg encoder",
.item_name = av_default_item_name,
.option = options,
.version = LIBAVUTIL_VERSION_INT,
};
AVCodec ff_mjpeg_encoder = {
.name = "mjpeg",
.long_name = NULL_IF_CONFIG_SMALL("MJPEG (Motion JPEG)"),
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_MJPEG,
.priv_data_size = sizeof(MpegEncContext),
.init = ff_mpv_encode_init,
.encode2 = ff_mpv_encode_picture,
.close = ff_mpv_encode_end,
.capabilities = AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_INTRA_ONLY,
.pix_fmts = (const enum AVPixelFormat[]){
AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_NONE
},
.priv_class = &mjpeg_class,
};
#endif
#if CONFIG_AMV_ENCODER
static const AVClass amv_class = {
.class_name = "amv encoder",
.item_name = av_default_item_name,
.option = options,
.version = LIBAVUTIL_VERSION_INT,
};
AVCodec ff_amv_encoder = {
.name = "amv",
.long_name = NULL_IF_CONFIG_SMALL("AMV Video"),
.type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_AMV,
.priv_data_size = sizeof(MpegEncContext),
.init = ff_mpv_encode_init,
.encode2 = amv_encode_picture,
.close = ff_mpv_encode_end,
.pix_fmts = (const enum AVPixelFormat[]){
AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_NONE
},
.priv_class = &amv_class,
};
#endif