summaryrefslogtreecommitdiff
path: root/src/libffmpeg/libavcodec/mpegvideo.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/libffmpeg/libavcodec/mpegvideo.c')
-rw-r--r--src/libffmpeg/libavcodec/mpegvideo.c2119
1 files changed, 1271 insertions, 848 deletions
diff --git a/src/libffmpeg/libavcodec/mpegvideo.c b/src/libffmpeg/libavcodec/mpegvideo.c
index 9f572c3d9..63242c9de 100644
--- a/src/libffmpeg/libavcodec/mpegvideo.c
+++ b/src/libffmpeg/libavcodec/mpegvideo.c
@@ -1,49 +1,42 @@
/*
* The simplest mpeg encoder (well, it was the simplest!)
- * Copyright (c) 2000,2001 Gerard Lantau.
+ * Copyright (c) 2000,2001 Fabrice Bellard.
*
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
+ * This library 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 of the License, or (at your option) any later version.
*
- * This program is distributed in the hope that it will be useful,
+ * This library 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 General Public License for more details.
+ * 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 General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
- * 4MV & hq encoding stuff by Michael Niedermayer <michaelni@gmx.at>
+ * 4MV & hq & b-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
*/
-#include <stdlib.h>
-#include <stdio.h>
-#include <math.h>
-#include <string.h>
#include "avcodec.h"
#include "dsputil.h"
#include "mpegvideo.h"
-#include "config.h"
-#include "xine-utils/xineutils.h"
-
#ifdef USE_FASTMEMCPY
#include "fastmemcpy.h"
#endif
static void encode_picture(MpegEncContext *s, int picture_number);
-static void rate_control_init(MpegEncContext *s);
-static int rate_estimate_qscale(MpegEncContext *s);
static void dct_unquantize_mpeg1_c(MpegEncContext *s,
DCTELEM *block, int n, int qscale);
+static void dct_unquantize_mpeg2_c(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale);
static void dct_unquantize_h263_c(MpegEncContext *s,
DCTELEM *block, int n, int qscale);
static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w);
-static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale);
+static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
-int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c;
+int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)= dct_quantize_c;
void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c;
#define EDGE_WIDTH 16
@@ -53,6 +46,7 @@ void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edg
//#define DEBUG
+
/* for jpeg fast DCT */
#define CONST_BITS 14
@@ -75,37 +69,55 @@ static UINT8 h263_chroma_roundtab[16] = {
static UINT16 default_mv_penalty[MAX_FCODE+1][MAX_MV*2+1];
static UINT8 default_fcode_tab[MAX_MV*2+1];
-/* default motion estimation */
-int motion_estimation_method = ME_LOG;
-
extern UINT8 zigzag_end[64];
-static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale)
+/* default motion estimation */
+int motion_estimation_method = ME_EPZS;
+
+static void convert_matrix(int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64],
+ const UINT16 *quant_matrix, int bias)
{
- int i;
+ int qscale;
- if (av_fdct == jpeg_fdct_ifast) {
- for(i=0;i<64;i++) {
- /* 16 <= qscale * quant_matrix[i] <= 7905 */
- /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
- /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
- /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
-
- qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
- (aanscales[i] * qscale * quant_matrix[block_permute_op(i)]));
- }
- } else {
- for(i=0;i<64;i++) {
- /* We can safely suppose that 16 <= quant_matrix[i] <= 255
- So 16 <= qscale * quant_matrix[i] <= 7905
- so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
- so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
- */
- qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
- qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
+ for(qscale=1; qscale<32; qscale++){
+ int i;
+ if (av_fdct == fdct_ifast) {
+ for(i=0;i<64;i++) {
+ const int j= block_permute_op(i);
+ /* 16 <= qscale * quant_matrix[i] <= 7905 */
+ /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */
+ /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
+ /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */
+
+ qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) /
+ (aanscales[i] * qscale * quant_matrix[j]));
+ }
+ } else {
+ for(i=0;i<64;i++) {
+ /* We can safely suppose that 16 <= quant_matrix[i] <= 255
+ So 16 <= qscale * quant_matrix[i] <= 7905
+ so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
+ so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67
+ */
+ qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
+ qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]);
+
+ if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1;
+
+ qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]);
+ }
}
}
}
+// move into common.c perhaps
+#define CHECKED_ALLOCZ(p, size)\
+{\
+ p= av_mallocz(size);\
+ if(p==NULL){\
+ perror("malloc");\
+ goto fail;\
+ }\
+}
/* init common structure for both encoder and decoder */
int MPV_common_init(MpegEncContext *s)
@@ -113,14 +125,19 @@ int MPV_common_init(MpegEncContext *s)
int c_size, i;
UINT8 *pict;
- if (s->out_format == FMT_H263)
- s->dct_unquantize = dct_unquantize_h263_c;
- else
- s->dct_unquantize = dct_unquantize_mpeg1_c;
+ s->dct_unquantize_h263 = dct_unquantize_h263_c;
+ s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_c;
+ s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_c;
#ifdef HAVE_MMX
MPV_common_init_mmx(s);
#endif
+ //setup default unquantizers (mpeg4 might change it later)
+ if(s->out_format == FMT_H263)
+ s->dct_unquantize = s->dct_unquantize_h263;
+ else
+ s->dct_unquantize = s->dct_unquantize_mpeg1;
+
s->mb_width = (s->width + 15) / 16;
s->mb_height = (s->height + 15) / 16;
s->mb_num = s->mb_width * s->mb_height;
@@ -135,58 +152,79 @@ int MPV_common_init(MpegEncContext *s)
c_size = (w >> shift) * (h >> shift);
pict_start = (w >> shift) * (EDGE_WIDTH >> shift) + (EDGE_WIDTH >> shift);
- pict = av_mallocz(c_size);
- if (pict == NULL)
- goto fail;
+ CHECKED_ALLOCZ(pict, c_size)
s->last_picture_base[i] = pict;
s->last_picture[i] = pict + pict_start;
+ if(i>0) memset(s->last_picture_base[i], 128, c_size);
- pict = av_mallocz(c_size);
- if (pict == NULL)
- goto fail;
+ CHECKED_ALLOCZ(pict, c_size)
s->next_picture_base[i] = pict;
s->next_picture[i] = pict + pict_start;
-
- if (s->has_b_frames) {
- pict = av_mallocz(c_size);
- if (pict == NULL)
- goto fail;
+ if(i>0) memset(s->next_picture_base[i], 128, c_size);
+
+ if (s->has_b_frames || s->codec_id==CODEC_ID_MPEG4) {
+ /* Note the MPEG4 stuff is here cuz of buggy encoders which dont set the low_delay flag but
+ do low-delay encoding, so we cant allways distinguish b-frame containing streams from low_delay streams */
+ CHECKED_ALLOCZ(pict, c_size)
s->aux_picture_base[i] = pict;
s->aux_picture[i] = pict + pict_start;
+ if(i>0) memset(s->aux_picture_base[i], 128, c_size);
}
}
if (s->encoding) {
- /* Allocate MB type table */
- s->mb_type = av_mallocz(s->mb_num * sizeof(char));
- if (s->mb_type == NULL) {
- perror("malloc");
- goto fail;
- }
+ int j;
+ int mv_table_size= (s->mb_width+2)*(s->mb_height+2);
+
+ CHECKED_ALLOCZ(s->mb_var , s->mb_num * sizeof(INT16))
+ CHECKED_ALLOCZ(s->mc_mb_var, s->mb_num * sizeof(INT16))
+
+ /* Allocate MV tables */
+ CHECKED_ALLOCZ(s->p_mv_table , mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_back_mv_table , mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_bidir_forw_mv_table , mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_bidir_back_mv_table , mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_direct_forw_mv_table, mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_direct_back_mv_table, mv_table_size * 2 * sizeof(INT16))
+ CHECKED_ALLOCZ(s->b_direct_mv_table , mv_table_size * 2 * sizeof(INT16))
+
+ CHECKED_ALLOCZ(s->me_scratchpad, s->linesize*16*3*sizeof(uint8_t))
- s->mb_var = av_mallocz(s->mb_num * sizeof(INT16));
- if (s->mb_var == NULL) {
- perror("malloc");
- goto fail;
+ CHECKED_ALLOCZ(s->me_map , ME_MAP_SIZE*sizeof(uint32_t))
+ CHECKED_ALLOCZ(s->me_score_map, ME_MAP_SIZE*sizeof(uint16_t))
+
+ if(s->max_b_frames){
+ for(j=0; j<REORDER_BUFFER_SIZE; j++){
+ int i;
+ for(i=0;i<3;i++) {
+ int w, h, shift;
+
+ w = s->linesize;
+ h = s->mb_height * 16;
+ shift = (i == 0) ? 0 : 1;
+ c_size = (w >> shift) * (h >> shift);
+
+ CHECKED_ALLOCZ(pict, c_size);
+ s->picture_buffer[j][i] = pict;
+ }
+ }
}
- /* Allocate MV table */
- /* By now we just have one MV per MB */
- s->mv_table[0] = av_mallocz(s->mb_num * sizeof(INT16));
- s->mv_table[1] = av_mallocz(s->mb_num * sizeof(INT16));
- if (s->mv_table[1] == NULL || s->mv_table[0] == NULL) {
- perror("malloc");
- goto fail;
+
+ if(s->codec_id==CODEC_ID_MPEG4){
+ CHECKED_ALLOCZ(s->tex_pb_buffer, PB_BUFFER_SIZE);
+ CHECKED_ALLOCZ( s->pb2_buffer, PB_BUFFER_SIZE);
}
}
if (s->out_format == FMT_H263 || s->encoding) {
int size;
+ /* Allocate MB type table */
+ CHECKED_ALLOCZ(s->mb_type , s->mb_num * sizeof(UINT8))
+
/* MV prediction */
size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
- s->motion_val = malloc(size * 2 * sizeof(INT16));
- if (s->motion_val == NULL)
- goto fail;
- memset(s->motion_val, 0, size * 2 * sizeof(INT16));
+ CHECKED_ALLOCZ(s->motion_val, size * 2 * sizeof(INT16));
}
if (s->h263_pred || s->h263_plus) {
@@ -197,43 +235,40 @@ int MPV_common_init(MpegEncContext *s)
y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
c_size = (s->mb_width + 2) * (s->mb_height + 2);
size = y_size + 2 * c_size;
- s->dc_val[0] = malloc(size * sizeof(INT16));
- if (s->dc_val[0] == NULL)
- goto fail;
+ CHECKED_ALLOCZ(s->dc_val[0], size * sizeof(INT16));
s->dc_val[1] = s->dc_val[0] + y_size;
s->dc_val[2] = s->dc_val[1] + c_size;
for(i=0;i<size;i++)
s->dc_val[0][i] = 1024;
/* ac values */
- s->ac_val[0] = av_mallocz(size * sizeof(INT16) * 16);
- if (s->ac_val[0] == NULL)
- goto fail;
+ CHECKED_ALLOCZ(s->ac_val[0], size * sizeof(INT16) * 16);
s->ac_val[1] = s->ac_val[0] + y_size;
s->ac_val[2] = s->ac_val[1] + c_size;
/* cbp values */
- s->coded_block = av_mallocz(y_size);
- if (!s->coded_block)
- goto fail;
+ CHECKED_ALLOCZ(s->coded_block, y_size);
/* which mb is a intra block */
- s->mbintra_table = av_mallocz(s->mb_num);
- if (!s->mbintra_table)
- goto fail;
+ CHECKED_ALLOCZ(s->mbintra_table, s->mb_num);
memset(s->mbintra_table, 1, s->mb_num);
+
+ /* divx501 bitstream reorder buffer */
+ CHECKED_ALLOCZ(s->bitstream_buffer, BITSTREAM_BUFFER_SIZE);
+
+ /* cbp, ac_pred, pred_dir */
+ CHECKED_ALLOCZ(s->cbp_table , s->mb_num * sizeof(UINT8))
+ CHECKED_ALLOCZ(s->pred_dir_table, s->mb_num * sizeof(UINT8))
+
+ CHECKED_ALLOCZ(s->qscale_table , s->mb_num * sizeof(UINT8))
}
/* default structure is frame */
s->picture_structure = PICT_FRAME;
/* init macroblock skip table */
- if (!s->encoding) {
- s->mbskip_table = av_mallocz(s->mb_num);
- if (!s->mbskip_table)
- goto fail;
- }
+ CHECKED_ALLOCZ(s->mbskip_table, s->mb_num);
- s->block= s->intra_block;
+ s->block= s->blocks[0];
s->context_initialized = 1;
return 0;
@@ -242,39 +277,49 @@ int MPV_common_init(MpegEncContext *s)
return -1;
}
+
+//extern int sads;
+
/* init common structure for both encoder and decoder */
void MPV_common_end(MpegEncContext *s)
{
int i;
- if (s->mb_type)
- free(s->mb_type);
- if (s->mb_var)
- free(s->mb_var);
- if (s->mv_table[0])
- free(s->mv_table[0]);
- if (s->mv_table[1])
- free(s->mv_table[1]);
- if (s->motion_val)
- free(s->motion_val);
- if (s->dc_val[0])
- free(s->dc_val[0]);
- if (s->ac_val[0])
- free(s->ac_val[0]);
- if (s->coded_block)
- free(s->coded_block);
- if (s->mbintra_table)
- free(s->mbintra_table);
-
- if (s->mbskip_table)
- free(s->mbskip_table);
+ av_freep(&s->mb_type);
+ av_freep(&s->mb_var);
+ av_freep(&s->mc_mb_var);
+ av_freep(&s->p_mv_table);
+ av_freep(&s->b_forw_mv_table);
+ av_freep(&s->b_back_mv_table);
+ av_freep(&s->b_bidir_forw_mv_table);
+ av_freep(&s->b_bidir_back_mv_table);
+ av_freep(&s->b_direct_forw_mv_table);
+ av_freep(&s->b_direct_back_mv_table);
+ av_freep(&s->b_direct_mv_table);
+ av_freep(&s->motion_val);
+ av_freep(&s->dc_val[0]);
+ av_freep(&s->ac_val[0]);
+ av_freep(&s->coded_block);
+ av_freep(&s->mbintra_table);
+ av_freep(&s->cbp_table);
+ av_freep(&s->pred_dir_table);
+ av_freep(&s->qscale_table);
+ av_freep(&s->me_scratchpad);
+ av_freep(&s->me_map);
+ av_freep(&s->me_score_map);
+
+ av_freep(&s->mbskip_table);
+ av_freep(&s->bitstream_buffer);
+ av_freep(&s->tex_pb_buffer);
+ av_freep(&s->pb2_buffer);
for(i=0;i<3;i++) {
- if (s->last_picture_base[i])
- free(s->last_picture_base[i]);
- if (s->next_picture_base[i])
- free(s->next_picture_base[i]);
- if (s->has_b_frames)
- free(s->aux_picture_base[i]);
+ int j;
+ av_freep(&s->last_picture_base[i]);
+ av_freep(&s->next_picture_base[i]);
+ av_freep(&s->aux_picture_base[i]);
+ for(j=0; j<REORDER_BUFFER_SIZE; j++){
+ av_freep(&s->picture_buffer[j][i]);
+ }
}
s->context_initialized = 0;
}
@@ -285,11 +330,17 @@ int MPV_encode_init(AVCodecContext *avctx)
MpegEncContext *s = avctx->priv_data;
int i;
+ avctx->pix_fmt = PIX_FMT_YUV420P;
+
s->bit_rate = avctx->bit_rate;
s->bit_rate_tolerance = avctx->bit_rate_tolerance;
s->frame_rate = avctx->frame_rate;
s->width = avctx->width;
s->height = avctx->height;
+ if(avctx->gop_size > 600){
+ fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n");
+ avctx->gop_size=600;
+ }
s->gop_size = avctx->gop_size;
s->rtp_mode = avctx->rtp_mode;
s->rtp_payload_size = avctx->rtp_payload_size;
@@ -300,36 +351,56 @@ int MPV_encode_init(AVCodecContext *avctx)
s->max_qdiff= avctx->max_qdiff;
s->qcompress= avctx->qcompress;
s->qblur= avctx->qblur;
+ s->b_quant_factor= avctx->b_quant_factor;
+ s->b_quant_offset= avctx->b_quant_offset;
s->avctx = avctx;
s->aspect_ratio_info= avctx->aspect_ratio_info;
s->flags= avctx->flags;
-
+ s->max_b_frames= avctx->max_b_frames;
+ s->rc_strategy= avctx->rc_strategy;
+ s->b_frame_strategy= avctx->b_frame_strategy;
+ s->codec_id= avctx->codec->id;
+ s->luma_elim_threshold = avctx->luma_elim_threshold;
+ s->chroma_elim_threshold= avctx->chroma_elim_threshold;
+ s->strict_std_compliance= avctx->strict_std_compliance;
+ s->data_partitioning= avctx->flags & CODEC_FLAG_PART;
+
if (s->gop_size <= 1) {
s->intra_only = 1;
s->gop_size = 12;
} else {
s->intra_only = 0;
}
- s->full_search = motion_estimation_method;
-
+
+ /* ME algorithm */
+ if (avctx->me_method == 0)
+ /* For compatibility */
+ s->me_method = motion_estimation_method;
+ else
+ s->me_method = avctx->me_method;
+
+ /* Fixed QSCALE */
s->fixed_qscale = (avctx->flags & CODEC_FLAG_QSCALE);
switch(avctx->codec->id) {
case CODEC_ID_MPEG1VIDEO:
s->out_format = FMT_MPEG1;
+ avctx->delay=0; //FIXME not sure, should check the spec
break;
case CODEC_ID_MJPEG:
s->out_format = FMT_MJPEG;
s->intra_only = 1; /* force intra only for jpeg */
s->mjpeg_write_tables = 1; /* write all tables */
+ s->mjpeg_data_only_frames = 0; /* write all the needed headers */
s->mjpeg_vsample[0] = 2; /* set up default sampling factors */
s->mjpeg_vsample[1] = 1; /* the only currently supported values */
s->mjpeg_vsample[2] = 1;
- s->mjpeg_hsample[0] = 2;
+ s->mjpeg_hsample[0] = 2;
s->mjpeg_hsample[1] = 1;
s->mjpeg_hsample[2] = 1;
if (mjpeg_init(s) < 0)
return -1;
+ avctx->delay=0;
break;
case CODEC_ID_H263:
if (h263_get_picture_format(s->width, s->height) == 7) {
@@ -337,6 +408,7 @@ int MPV_encode_init(AVCodecContext *avctx)
return -1;
}
s->out_format = FMT_H263;
+ avctx->delay=0;
break;
case CODEC_ID_H263P:
s->out_format = FMT_H263;
@@ -344,19 +416,25 @@ int MPV_encode_init(AVCodecContext *avctx)
s->rtp_payload_size = 1200;
s->h263_plus = 1;
s->unrestricted_mv = 1;
+ s->h263_aic = 1;
/* These are just to be sure */
s->umvplus = 0;
s->umvplus_dec = 0;
+ avctx->delay=0;
break;
case CODEC_ID_RV10:
s->out_format = FMT_H263;
s->h263_rv10 = 1;
+ avctx->delay=0;
break;
case CODEC_ID_MPEG4:
s->out_format = FMT_H263;
s->h263_pred = 1;
s->unrestricted_mv = 1;
+ s->has_b_frames= s->max_b_frames ? 1 : 0;
+ s->low_delay=0;
+ avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
break;
case CODEC_ID_MSMPEG4V1:
s->out_format = FMT_H263;
@@ -364,6 +442,7 @@ int MPV_encode_init(AVCodecContext *avctx)
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 1;
+ avctx->delay=0;
break;
case CODEC_ID_MSMPEG4V2:
s->out_format = FMT_H263;
@@ -371,6 +450,7 @@ int MPV_encode_init(AVCodecContext *avctx)
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 2;
+ avctx->delay=0;
break;
case CODEC_ID_MSMPEG4V3:
s->out_format = FMT_H263;
@@ -378,16 +458,12 @@ int MPV_encode_init(AVCodecContext *avctx)
s->h263_pred = 1;
s->unrestricted_mv = 1;
s->msmpeg4_version= 3;
+ avctx->delay=0;
break;
default:
return -1;
}
- if((s->flags&CODEC_FLAG_4MV) && !(s->flags&CODEC_FLAG_HQ)){
- printf("4MV is currently only supported in HQ mode\n");
- return -1;
- }
-
{ /* set up some save defaults, some codecs might override them later */
static int done=0;
if(!done){
@@ -410,7 +486,7 @@ int MPV_encode_init(AVCodecContext *avctx)
mpeg1_encode_init(s);
/* dont use mv_penalty table for crap MV as it would be confused */
- if(s->full_search<4) s->mv_penalty= default_mv_penalty;
+ if (s->me_method < ME_EPZS) s->mv_penalty = default_mv_penalty;
s->encoding = 1;
@@ -420,18 +496,32 @@ int MPV_encode_init(AVCodecContext *avctx)
/* init default q matrix */
for(i=0;i<64;i++) {
- s->intra_matrix[i] = default_intra_matrix[i];
- s->non_intra_matrix[i] = default_non_intra_matrix[i];
+ if(s->out_format == FMT_H263)
+ s->intra_matrix[i] = default_non_intra_matrix[i];
+ else
+ s->intra_matrix[i] = default_intra_matrix[i];
+
+ s->inter_matrix[i] = default_non_intra_matrix[i];
}
- /* rate control init */
- rate_control_init(s);
+ /* precompute matrix */
+ /* for mjpeg, we do include qscale in the matrix */
+ if (s->out_format != FMT_MJPEG) {
+ convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias,
+ s->intra_matrix, s->intra_quant_bias);
+ convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias,
+ s->inter_matrix, s->inter_quant_bias);
+ }
+
+ if(ff_rate_control_init(s) < 0)
+ return -1;
s->picture_number = 0;
s->picture_in_gop_number = 0;
s->fake_picture_number = 0;
/* motion detector init */
s->f_code = 1;
+ s->b_code = 1;
return 0;
}
@@ -443,6 +533,9 @@ int MPV_encode_end(AVCodecContext *avctx)
#ifdef STATS
print_stats();
#endif
+
+ ff_rate_control_uninit(s);
+
MPV_common_end(s);
if (s->out_format == FMT_MJPEG)
mjpeg_close(s);
@@ -485,12 +578,13 @@ void MPV_frame_start(MpegEncContext *s)
UINT8 *tmp;
s->mb_skiped = 0;
+ s->decoding_error=0;
+
if (s->pict_type == B_TYPE) {
for(i=0;i<3;i++) {
s->current_picture[i] = s->aux_picture[i];
}
} else {
- s->last_non_b_pict_type= s->pict_type;
for(i=0;i<3;i++) {
/* swap next and last */
tmp = s->last_picture[i];
@@ -504,9 +598,11 @@ void MPV_frame_start(MpegEncContext *s)
/* generic function for encode/decode called after a frame has been coded/decoded */
void MPV_frame_end(MpegEncContext *s)
{
+// if((s->picture_number%100)==0 && s->encoding) printf("sads:%d //\n", sads);
+
/* draw edge for correct motion prediction if outside */
if (s->pict_type != B_TYPE && !s->intra_only) {
- if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version==500){
+ if(s->avctx==NULL || s->avctx->codec->id!=CODEC_ID_MPEG4 || s->divx_version>=500){
draw_edges(s->current_picture[0], s->linesize, s->mb_width*16, s->mb_height*16, EDGE_WIDTH);
draw_edges(s->current_picture[1], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
draw_edges(s->current_picture[2], s->linesize/2, s->mb_width*8, s->mb_height*8, EDGE_WIDTH/2);
@@ -518,107 +614,195 @@ void MPV_frame_end(MpegEncContext *s)
}
}
emms_c();
+
+ if(s->pict_type!=B_TYPE){
+ s->last_non_b_pict_type= s->pict_type;
+ s->last_non_b_qscale= s->qscale;
+ s->last_non_b_mc_mb_var= s->mc_mb_var_sum;
+ s->num_available_buffers++;
+ if(s->num_available_buffers>2) s->num_available_buffers= 2;
+ }
}
-int MPV_encode_picture(AVCodecContext *avctx,
- unsigned char *buf, int buf_size, void *data)
+/* reorder input for encoding */
+void reorder_input(MpegEncContext *s, AVPicture *pict)
{
- MpegEncContext *s = avctx->priv_data;
- AVPicture *pict = data;
- int i, j;
+ int i, j, index;
+
+ if(s->max_b_frames > FF_MAX_B_FRAMES) s->max_b_frames= FF_MAX_B_FRAMES;
- if (s->fixed_qscale)
- s->qscale = avctx->quality;
+// delay= s->max_b_frames+1; (or 0 if no b frames cuz decoder diff)
- init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
-
- s->force_type= (avctx->flags&CODEC_FLAG_TYPE) ?
- (avctx->key_frame ? I_TYPE : P_TYPE) : 0;
- if (!s->intra_only) {
- /* first picture of GOP is intra */
- if (s->picture_in_gop_number % s->gop_size==0 || s->force_type==I_TYPE){
- s->picture_in_gop_number=0;
- s->pict_type = I_TYPE;
- }else
- s->pict_type = P_TYPE;
- } else {
- s->pict_type = I_TYPE;
+ for(j=0; j<REORDER_BUFFER_SIZE-1; j++){
+ s->coded_order[j]= s->coded_order[j+1];
}
-
- MPV_frame_start(s);
-
- for(i=0;i<3;i++) {
- UINT8 *src = pict->data[i];
- UINT8 *dest = s->current_picture[i];
- int src_wrap = pict->linesize[i];
- int dest_wrap = s->linesize;
- int w = s->width;
- int h = s->height;
-
- if (i >= 1) {
- dest_wrap >>= 1;
- w >>= 1;
- h >>= 1;
+ s->coded_order[j].picture[0]= s->coded_order[j].picture[1]= s->coded_order[j].picture[2]= NULL; //catch uninitalized buffers
+ s->coded_order[j].pict_type=0;
+
+ switch(s->input_pict_type){
+ default:
+ case I_TYPE:
+ case S_TYPE:
+ case P_TYPE:
+ index= s->max_b_frames - s->b_frames_since_non_b;
+ s->b_frames_since_non_b=0;
+ break;
+ case B_TYPE:
+ index= s->max_b_frames + 1;
+ s->b_frames_since_non_b++;
+ break;
+ }
+//printf("index:%d type:%d strides: %d %d\n", index, s->input_pict_type, pict->linesize[0], s->linesize);
+ if( (index==0 || (s->flags&CODEC_FLAG_INPUT_PRESERVED))
+ && pict->linesize[0] == s->linesize
+ && pict->linesize[1] == s->linesize>>1
+ && pict->linesize[2] == s->linesize>>1){
+//printf("ptr\n");
+ for(i=0; i<3; i++){
+ s->coded_order[index].picture[i]= pict->data[i];
}
+ }else{
+//printf("copy\n");
+ for(i=0; i<3; i++){
+ uint8_t *src = pict->data[i];
+ uint8_t *dest;
+ int src_wrap = pict->linesize[i];
+ int dest_wrap = s->linesize;
+ int w = s->width;
+ int h = s->height;
+
+ if(index==0) dest= s->last_picture[i]+16; //is current_picture indeed but the switch hapens after reordering
+ else dest= s->picture_buffer[s->picture_buffer_index][i];
+
+ if (i >= 1) {
+ dest_wrap >>= 1;
+ w >>= 1;
+ h >>= 1;
+ }
- if(dest_wrap==src_wrap){
- s->new_picture[i] = pict->data[i];
- } else {
+ s->coded_order[index].picture[i]= dest;
for(j=0;j<h;j++) {
memcpy(dest, src, w);
dest += dest_wrap;
src += src_wrap;
}
- s->new_picture[i] = s->current_picture[i];
- }
+ }
+ if(index!=0){
+ s->picture_buffer_index++;
+ if(s->picture_buffer_index >= REORDER_BUFFER_SIZE-1) s->picture_buffer_index=0;
+ }
+ }
+ s->coded_order[index].pict_type = s->input_pict_type;
+ s->coded_order[index].qscale = s->input_qscale;
+ s->coded_order[index].force_type= s->force_input_type;
+ s->coded_order[index].picture_in_gop_number= s->input_picture_in_gop_number;
+ s->coded_order[index].picture_number= s->input_picture_number;
+
+ for(i=0; i<3; i++){
+ s->new_picture[i]= s->coded_order[0].picture[i];
}
+}
+
+int MPV_encode_picture(AVCodecContext *avctx,
+ unsigned char *buf, int buf_size, void *data)
+{
+ MpegEncContext *s = avctx->priv_data;
+ AVPicture *pict = data;
- encode_picture(s, s->picture_number);
- avctx->key_frame = (s->pict_type == I_TYPE);
- avctx->header_bits = s->header_bits;
- avctx->mv_bits = s->mv_bits;
- avctx->misc_bits = s->misc_bits;
- avctx->i_tex_bits = s->i_tex_bits;
- avctx->p_tex_bits = s->p_tex_bits;
- avctx->i_count = s->i_count;
- avctx->p_count = s->p_count;
- avctx->skip_count = s->skip_count;
+ s->input_qscale = avctx->quality;
- MPV_frame_end(s);
- s->picture_number++;
- s->picture_in_gop_number++;
+ init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
- if (s->out_format == FMT_MJPEG)
- mjpeg_picture_trailer(s);
+ if(avctx->flags&CODEC_FLAG_TYPE){
+ s->input_pict_type=
+ s->force_input_type= avctx->key_frame ? I_TYPE : P_TYPE;
+ }else if(s->flags&CODEC_FLAG_PASS2){
+ s->input_pict_type=
+ s->force_input_type= s->rc_context.entry[s->input_picture_number].new_pict_type;
+ }else{
+ s->force_input_type=0;
+ if (!s->intra_only) {
+ /* first picture of GOP is intra */
+ if (s->input_picture_in_gop_number % s->gop_size==0){
+ s->input_pict_type = I_TYPE;
+ }else if(s->max_b_frames==0){
+ s->input_pict_type = P_TYPE;
+ }else{
+ if(s->b_frames_since_non_b < s->max_b_frames) //FIXME more IQ
+ s->input_pict_type = B_TYPE;
+ else
+ s->input_pict_type = P_TYPE;
+ }
+ } else {
+ s->input_pict_type = I_TYPE;
+ }
+ }
+
+ if(s->input_pict_type==I_TYPE)
+ s->input_picture_in_gop_number=0;
+
+ reorder_input(s, pict);
+
+ /* output? */
+ if(s->coded_order[0].picture[0]){
+
+ s->pict_type= s->coded_order[0].pict_type;
+ if (s->fixed_qscale) /* the ratecontrol needs the last qscale so we dont touch it for CBR */
+ s->qscale= s->coded_order[0].qscale;
+ s->force_type= s->coded_order[0].force_type;
+ s->picture_in_gop_number= s->coded_order[0].picture_in_gop_number;
+ s->picture_number= s->coded_order[0].picture_number;
+
+ MPV_frame_start(s);
+
+ encode_picture(s, s->picture_number);
+ avctx->key_frame = (s->pict_type == I_TYPE);
+ avctx->pict_type = s->pict_type;
+ avctx->real_pict_num = s->picture_number;
+ avctx->header_bits = s->header_bits;
+ avctx->mv_bits = s->mv_bits;
+ avctx->misc_bits = s->misc_bits;
+ avctx->i_tex_bits = s->i_tex_bits;
+ avctx->p_tex_bits = s->p_tex_bits;
+ avctx->i_count = s->i_count;
+ avctx->p_count = s->p_count;
+ avctx->skip_count = s->skip_count;
+
+ MPV_frame_end(s);
+
+ if (s->out_format == FMT_MJPEG)
+ mjpeg_picture_trailer(s);
+
+ avctx->quality = s->qscale;
+
+ if(s->flags&CODEC_FLAG_PASS1)
+ ff_write_pass1_stats(s);
+
+ }
+
+ s->input_picture_number++;
+ s->input_picture_in_gop_number++;
flush_put_bits(&s->pb);
- s->last_frame_bits= s->frame_bits;
s->frame_bits = (pbBufPtr(&s->pb) - s->pb.buf) * 8;
+ if(s->pict_type==B_TYPE) s->pb_frame_bits+= s->frame_bits;
+ else s->pb_frame_bits= s->frame_bits;
+
s->total_bits += s->frame_bits;
avctx->frame_bits = s->frame_bits;
//printf("fcode: %d, type: %d, head: %d, mv: %d, misc: %d, frame: %d, itex: %d, ptex: %d\n",
//s->f_code, avctx->key_frame, s->header_bits, s->mv_bits, s->misc_bits, s->frame_bits, s->i_tex_bits, s->p_tex_bits);
- avctx->quality = s->qscale;
if (avctx->get_psnr) {
/* At this point pict->data should have the original frame */
/* an s->current_picture should have the coded/decoded frame */
get_psnr(pict->data, s->current_picture,
pict->linesize, s->linesize, avctx);
+// printf("%f\n", avctx->psnr_y);
}
return pbBufPtr(&s->pb) - s->pb.buf;
}
-static inline int clip(int a, int amin, int amax)
-{
- if (a < amin)
- return amin;
- else if (a > amax)
- return amax;
- else
- return a;
-}
-
static inline void gmc1_motion(MpegEncContext *s,
UINT8 *dest_y, UINT8 *dest_cb, UINT8 *dest_cr,
int dest_offset,
@@ -626,7 +810,7 @@ static inline void gmc1_motion(MpegEncContext *s,
int h)
{
UINT8 *ptr;
- int dxy, offset, mx, my, src_x, src_y, height, linesize;
+ int offset, src_x, src_y, linesize;
int motion_x, motion_y;
if(s->real_sprite_warping_points>1) printf("more than 1 warp point isnt supported\n");
@@ -705,6 +889,8 @@ if(s->quarter_sample)
pix_op[dxy](dest_y, ptr, linesize, h);
pix_op[dxy](dest_y + 8, ptr + 8, linesize, h);
+ if(s->flags&CODEC_FLAG_GRAY) return;
+
if (s->out_format == FMT_H263) {
dxy = 0;
if ((motion_x & 3) != 0)
@@ -768,6 +954,8 @@ static inline void qpel_motion(MpegEncContext *s,
qpix_op[dxy](dest_y + linesize*8 , ptr + linesize*8 , linesize, linesize, motion_x&3, motion_y&3);
qpix_op[dxy](dest_y + linesize*8 + 8, ptr + linesize*8 + 8, linesize, linesize, motion_x&3, motion_y&3);
+ if(s->flags&CODEC_FLAG_GRAY) return;
+
mx= (motion_x>>1) | (motion_x&1);
my= (motion_y>>1) | (motion_y&1);
@@ -856,6 +1044,8 @@ static inline void MPV_motion(MpegEncContext *s,
dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
pix_op[dxy](dest, ptr, s->linesize, 8);
}
+
+ if(s->flags&CODEC_FLAG_GRAY) break;
/* In case of 8X8, we construct a single chroma motion vector
with a special rounding */
mx = 0;
@@ -922,8 +1112,7 @@ static inline void put_dct(MpegEncContext *s,
{
if (!s->mpeg2)
s->dct_unquantize(s, block, i, s->qscale);
- ff_idct (block);
- put_pixels_clamped(block, dest, line_size);
+ ff_idct_put (dest, line_size, block);
}
/* add block[] to dest[] */
@@ -931,14 +1120,53 @@ static inline void add_dct(MpegEncContext *s,
DCTELEM *block, int i, UINT8 *dest, int line_size)
{
if (s->block_last_index[i] >= 0) {
- if (!s->mpeg2)
- if(s->encoding || (!s->h263_msmpeg4))
- s->dct_unquantize(s, block, i, s->qscale);
- ff_idct (block);
- add_pixels_clamped(block, dest, line_size);
+ ff_idct_add (dest, line_size, block);
}
}
+static inline void add_dequant_dct(MpegEncContext *s,
+ DCTELEM *block, int i, UINT8 *dest, int line_size)
+{
+ if (s->block_last_index[i] >= 0) {
+ s->dct_unquantize(s, block, i, s->qscale);
+
+ ff_idct_add (dest, line_size, block);
+ }
+}
+
+/**
+ * cleans dc, ac, coded_block for the current non intra MB
+ */
+void ff_clean_intra_table_entries(MpegEncContext *s)
+{
+ int wrap = s->block_wrap[0];
+ int xy = s->block_index[0];
+
+ s->dc_val[0][xy ] =
+ s->dc_val[0][xy + 1 ] =
+ s->dc_val[0][xy + wrap] =
+ s->dc_val[0][xy + 1 + wrap] = 1024;
+ /* ac pred */
+ memset(s->ac_val[0][xy ], 0, 32 * sizeof(INT16));
+ memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(INT16));
+ if (s->msmpeg4_version>=3) {
+ s->coded_block[xy ] =
+ s->coded_block[xy + 1 ] =
+ s->coded_block[xy + wrap] =
+ s->coded_block[xy + 1 + wrap] = 0;
+ }
+ /* chroma */
+ wrap = s->block_wrap[4];
+ xy = s->mb_x + 1 + (s->mb_y + 1) * wrap;
+ s->dc_val[1][xy] =
+ s->dc_val[2][xy] = 1024;
+ /* ac pred */
+ memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
+ memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
+
+ s->mbintra_table[s->mb_x + s->mb_y*s->mb_width]= 0;
+}
+
/* generic function called after a macroblock has been parsed by the
decoder or after it has been encoded by the encoder.
@@ -952,69 +1180,39 @@ static inline void add_dct(MpegEncContext *s,
void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
{
int mb_x, mb_y;
- int dct_linesize, dct_offset;
- op_pixels_func *op_pix;
- qpel_mc_func *op_qpix;
+ const int mb_xy = s->mb_y * s->mb_width + s->mb_x;
mb_x = s->mb_x;
mb_y = s->mb_y;
#ifdef FF_POSTPROCESS
+ /* Obsolete. Exists for compatibility with mplayer only. */
quant_store[mb_y][mb_x]=s->qscale;
//printf("[%02d][%02d] %d\n",mb_x,mb_y,s->qscale);
+#else
+ if(s->avctx->quant_store) s->avctx->quant_store[mb_y*s->avctx->qstride+mb_x] = s->qscale;
#endif
/* update DC predictors for P macroblocks */
if (!s->mb_intra) {
if (s->h263_pred || s->h263_aic) {
- if(s->mbintra_table[mb_x + mb_y*s->mb_width])
- {
- int wrap, xy, v;
- s->mbintra_table[mb_x + mb_y*s->mb_width]=0;
- wrap = 2 * s->mb_width + 2;
- xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
- v = 1024;
-
- s->dc_val[0][xy] = v;
- s->dc_val[0][xy + 1] = v;
- s->dc_val[0][xy + wrap] = v;
- s->dc_val[0][xy + 1 + wrap] = v;
- /* ac pred */
- memset(s->ac_val[0][xy], 0, 16 * sizeof(INT16));
- memset(s->ac_val[0][xy + 1], 0, 16 * sizeof(INT16));
- memset(s->ac_val[0][xy + wrap], 0, 16 * sizeof(INT16));
- memset(s->ac_val[0][xy + 1 + wrap], 0, 16 * sizeof(INT16));
- if (s->h263_msmpeg4) {
- s->coded_block[xy] = 0;
- s->coded_block[xy + 1] = 0;
- s->coded_block[xy + wrap] = 0;
- s->coded_block[xy + 1 + wrap] = 0;
- }
- /* chroma */
- wrap = s->mb_width + 2;
- xy = mb_x + 1 + (mb_y + 1) * wrap;
- s->dc_val[1][xy] = v;
- s->dc_val[2][xy] = v;
- /* ac pred */
- memset(s->ac_val[1][xy], 0, 16 * sizeof(INT16));
- memset(s->ac_val[2][xy], 0, 16 * sizeof(INT16));
- }
+ if(s->mbintra_table[mb_xy])
+ ff_clean_intra_table_entries(s);
} else {
- s->last_dc[0] = 128 << s->intra_dc_precision;
- s->last_dc[1] = 128 << s->intra_dc_precision;
+ s->last_dc[0] =
+ s->last_dc[1] =
s->last_dc[2] = 128 << s->intra_dc_precision;
}
}
else if (s->h263_pred || s->h263_aic)
- s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
+ s->mbintra_table[mb_xy]=1;
/* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
- if (s->out_format == FMT_H263) {
- if(s->pict_type!=B_TYPE){
- int xy, wrap, motion_x, motion_y;
+ if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
+ int motion_x, motion_y;
- wrap = 2 * s->mb_width + 2;
- xy = 2 * mb_x + 1 + (2 * mb_y + 1) * wrap;
+ const int wrap = s->block_wrap[0];
+ const int xy = s->block_index[0];
if (s->mb_intra) {
motion_x = 0;
motion_y = 0;
@@ -1033,20 +1231,23 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
s->motion_val[xy + 1 + wrap][0] = motion_x;
s->motion_val[xy + 1 + wrap][1] = motion_y;
}
- }
}
- if (!s->intra_only) {
+ if (!(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) {
UINT8 *dest_y, *dest_cb, *dest_cr;
- UINT8 *mbskip_ptr;
-
- /* avoid copy if macroblock skipped in last frame too */
- if (!s->encoding && s->pict_type != B_TYPE) {
- mbskip_ptr = &s->mbskip_table[s->mb_y * s->mb_width + s->mb_x];
+ int dct_linesize, dct_offset;
+ op_pixels_func *op_pix;
+ qpel_mc_func *op_qpix;
+
+ /* avoid copy if macroblock skipped in last frame too
+ dont touch it for B-frames as they need the skip info from the next p-frame */
+ if (s->pict_type != B_TYPE) {
+ UINT8 *mbskip_ptr = &s->mbskip_table[mb_xy];
if (s->mb_skiped) {
s->mb_skiped = 0;
- /* if previous was skipped too, then nothing to do ! */
- if (*mbskip_ptr != 0)
+ /* if previous was skipped too, then nothing to do !
+ skip only during decoding as we might trash the buffers during encoding a bit */
+ if (*mbskip_ptr != 0 && !s->encoding)
goto the_end;
*mbskip_ptr = 1; /* indicate that this time we skiped it */
} else {
@@ -1068,33 +1269,53 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
if (!s->mb_intra) {
/* motion handling */
- if (!s->no_rounding){
- op_pix = put_pixels_tab;
- op_qpix= qpel_mc_rnd_tab;
- }else{
- op_pix = put_no_rnd_pixels_tab;
- op_qpix= qpel_mc_no_rnd_tab;
- }
+ /* decoding or more than one mb_type (MC was allready done otherwise) */
+ if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
+ if ((!s->no_rounding) || s->pict_type==B_TYPE){
+ op_pix = put_pixels_tab;
+ op_qpix= qpel_mc_rnd_tab;
+ }else{
+ op_pix = put_no_rnd_pixels_tab;
+ op_qpix= qpel_mc_no_rnd_tab;
+ }
- if (s->mv_dir & MV_DIR_FORWARD) {
- MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
- if (!s->no_rounding)
- op_pix = avg_pixels_tab;
- else
- op_pix = avg_no_rnd_pixels_tab;
- }
- if (s->mv_dir & MV_DIR_BACKWARD) {
- MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
+ if (s->mv_dir & MV_DIR_FORWARD) {
+ MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
+ if ((!s->no_rounding) || s->pict_type==B_TYPE)
+ op_pix = avg_pixels_tab;
+ else
+ op_pix = avg_no_rnd_pixels_tab;
+ }
+ if (s->mv_dir & MV_DIR_BACKWARD) {
+ MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
+ }
}
- /* add dct residue */
- add_dct(s, block[0], 0, dest_y, dct_linesize);
- add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
- add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
- add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
+ /* skip dequant / idct if we are really late ;) */
+ if(s->hurry_up>1) goto the_end;
- add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
- add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
+ /* add dct residue */
+ if(!s->mpeg2 && (s->encoding || (!s->h263_msmpeg4))){
+ add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
+ add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
+ add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
+ add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
+
+ if(!(s->flags&CODEC_FLAG_GRAY)){
+ add_dequant_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
+ add_dequant_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
+ }
+ } else {
+ add_dct(s, block[0], 0, dest_y, dct_linesize);
+ add_dct(s, block[1], 1, dest_y + 8, dct_linesize);
+ add_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
+ add_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
+
+ if(!(s->flags&CODEC_FLAG_GRAY)){
+ add_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
+ add_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
+ }
+ }
} else {
/* dct only in intra block */
put_dct(s, block[0], 0, dest_y, dct_linesize);
@@ -1102,128 +1323,188 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
put_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
- put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
- put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
+ if(!(s->flags&CODEC_FLAG_GRAY)){
+ put_dct(s, block[4], 4, dest_cb, s->linesize >> 1);
+ put_dct(s, block[5], 5, dest_cr, s->linesize >> 1);
+ }
}
}
the_end:
emms_c(); //FIXME remove
}
-static void encode_mb(MpegEncContext *s)
+static inline void dct_single_coeff_elimination(MpegEncContext *s, int n, int threshold, int skip_dc)
+{
+ static const char tab[64]=
+ {3,2,2,1,1,1,1,1,
+ 1,1,1,1,1,1,1,1,
+ 1,1,1,1,1,1,1,1,
+ 0,0,0,0,0,0,0,0,
+ 0,0,0,0,0,0,0,0,
+ 0,0,0,0,0,0,0,0,
+ 0,0,0,0,0,0,0,0,
+ 0,0,0,0,0,0,0,0};
+ int score=0;
+ int run=0;
+ int i;
+ DCTELEM *block= s->block[n];
+ const int last_index= s->block_last_index[n];
+
+ if(skip_dc) skip_dc=1;
+
+ /* are all which we could set to zero are allready zero? */
+ if(last_index<=skip_dc - 1) return;
+
+ for(i=0; i<=last_index; i++){
+ const int j = zigzag_direct[i];
+ const int level = ABS(block[j]);
+ if(level==1){
+ if(skip_dc && i==0) continue;
+ score+= tab[run];
+ run=0;
+ }else if(level>1){
+ return;
+ }else{
+ run++;
+ }
+ }
+ if(score >= threshold) return;
+ for(i=skip_dc; i<=last_index; i++){
+ const int j = zigzag_direct[i];
+ block[j]=0;
+ }
+ if(block[0]) s->block_last_index[n]= 0;
+ else s->block_last_index[n]= -1;
+}
+
+static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index)
+{
+ int i;
+ const int maxlevel= s->max_qcoeff;
+ const int minlevel= s->min_qcoeff;
+
+ for(i=0;i<=last_index; i++){
+ const int j = zigzag_direct[i];
+ int level = block[j];
+
+ if (level>maxlevel) level=maxlevel;
+ else if(level<minlevel) level=minlevel;
+ block[j]= level;
+ }
+}
+
+static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
{
- int wrap;
const int mb_x= s->mb_x;
const int mb_y= s->mb_y;
- UINT8 *ptr;
- const int motion_x= s->mv[0][0][0];
- const int motion_y= s->mv[0][0][1];
int i;
+ int skip_dct[6];
+#if 0
+ if (s->interlaced_dct) {
+ dct_linesize = s->linesize * 2;
+ dct_offset = s->linesize;
+ } else {
+ dct_linesize = s->linesize;
+ dct_offset = s->linesize * 8;
+ }
+#endif
+ for(i=0; i<6; i++) skip_dct[i]=0;
- /* get the pixels */
- wrap = s->linesize;
- ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
- get_pixels(s->block[0], ptr, wrap);
- get_pixels(s->block[1], ptr + 8, wrap);
- get_pixels(s->block[2], ptr + 8 * wrap, wrap);
- get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
- wrap = s->linesize >> 1;
- ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
- get_pixels(s->block[4], ptr, wrap);
-
- wrap = s->linesize >> 1;
- ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
- get_pixels(s->block[5], ptr, wrap);
-
- /* subtract previous frame if non intra */
- if (!s->mb_intra) {
- int dxy, offset, mx, my;
+ if (s->mb_intra) {
+ UINT8 *ptr;
+ int wrap;
+
+ wrap = s->linesize;
+ ptr = s->new_picture[0] + (mb_y * 16 * wrap) + mb_x * 16;
+ get_pixels(s->block[0], ptr , wrap);
+ get_pixels(s->block[1], ptr + 8, wrap);
+ get_pixels(s->block[2], ptr + 8 * wrap , wrap);
+ get_pixels(s->block[3], ptr + 8 * wrap + 8, wrap);
+
+ if(s->flags&CODEC_FLAG_GRAY){
+ skip_dct[4]= 1;
+ skip_dct[5]= 1;
+ }else{
+ wrap >>=1;
+ ptr = s->new_picture[1] + (mb_y * 8 * wrap) + mb_x * 8;
+ get_pixels(s->block[4], ptr, wrap);
+
+ ptr = s->new_picture[2] + (mb_y * 8 * wrap) + mb_x * 8;
+ get_pixels(s->block[5], ptr, wrap);
+ }
+ }else{
+ op_pixels_func *op_pix;
+ qpel_mc_func *op_qpix;
+ UINT8 *dest_y, *dest_cb, *dest_cr;
+ UINT8 *ptr_y, *ptr_cb, *ptr_cr;
+ int wrap_y, wrap_c;
+
+ dest_y = s->current_picture[0] + (mb_y * 16 * s->linesize ) + mb_x * 16;
+ dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
+ dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
+ wrap_y = s->linesize;
+ wrap_c = wrap_y>>1;
+ ptr_y = s->new_picture[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
+ ptr_cb = s->new_picture[1] + (mb_y * 8 * wrap_c) + mb_x * 8;
+ ptr_cr = s->new_picture[2] + (mb_y * 8 * wrap_c) + mb_x * 8;
+
+ if ((!s->no_rounding) || s->pict_type==B_TYPE){
+ op_pix = put_pixels_tab;
+ op_qpix= qpel_mc_rnd_tab;
+ }else{
+ op_pix = put_no_rnd_pixels_tab;
+ op_qpix= qpel_mc_no_rnd_tab;
+ }
+
+ if (s->mv_dir & MV_DIR_FORWARD) {
+ MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture, op_pix, op_qpix);
+ if ((!s->no_rounding) || s->pict_type==B_TYPE)
+ op_pix = avg_pixels_tab;
+ else
+ op_pix = avg_no_rnd_pixels_tab;
+ }
+ if (s->mv_dir & MV_DIR_BACKWARD) {
+ MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture, op_pix, op_qpix);
+ }
+
+ diff_pixels(s->block[0], ptr_y , dest_y , wrap_y);
+ diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
+ diff_pixels(s->block[2], ptr_y + 8 * wrap_y , dest_y + 8 * wrap_y , wrap_y);
+ diff_pixels(s->block[3], ptr_y + 8 * wrap_y + 8, dest_y + 8 * wrap_y + 8, wrap_y);
- if(s->mv_type==MV_TYPE_16X16){
- dxy = ((motion_y & 1) << 1) | (motion_x & 1);
- ptr = s->last_picture[0] +
- ((mb_y * 16 + (motion_y >> 1)) * s->linesize) +
- (mb_x * 16 + (motion_x >> 1));
-
- sub_pixels_2(s->block[0], ptr, s->linesize, dxy);
- sub_pixels_2(s->block[1], ptr + 8, s->linesize, dxy);
- sub_pixels_2(s->block[2], ptr + s->linesize * 8, s->linesize, dxy);
- sub_pixels_2(s->block[3], ptr + 8 + s->linesize * 8, s->linesize ,dxy);
-
- if (s->out_format == FMT_H263) {
- /* special rounding for h263 */
- dxy = 0;
- if ((motion_x & 3) != 0)
- dxy |= 1;
- if ((motion_y & 3) != 0)
- dxy |= 2;
- mx = motion_x >> 2;
- my = motion_y >> 2;
- } else {
- mx = motion_x / 2;
- my = motion_y / 2;
- dxy = ((my & 1) << 1) | (mx & 1);
- mx >>= 1;
- my >>= 1;
- }
- offset = ((mb_y * 8 + my) * (s->linesize >> 1)) + (mb_x * 8 + mx);
- ptr = s->last_picture[1] + offset;
- sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy);
- ptr = s->last_picture[2] + offset;
- sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy);
+ if(s->flags&CODEC_FLAG_GRAY){
+ skip_dct[4]= 1;
+ skip_dct[5]= 1;
}else{
- int src_x, src_y;
-
- for(i=0;i<4;i++) {
- int motion_x = s->mv[0][i][0];
- int motion_y = s->mv[0][i][1];
-
- dxy = ((motion_y & 1) << 1) | (motion_x & 1);
- src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
- src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
-
- ptr = s->last_picture[0] + (src_y * s->linesize) + (src_x);
- sub_pixels_2(s->block[i], ptr, s->linesize, dxy);
- }
- /* In case of 8X8, we construct a single chroma motion vector
- with a special rounding */
- mx = 0;
- my = 0;
- for(i=0;i<4;i++) {
- mx += s->mv[0][i][0];
- my += s->mv[0][i][1];
- }
- if (mx >= 0)
- mx = (h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
- else {
- mx = -mx;
- mx = -(h263_chroma_roundtab[mx & 0xf] + ((mx >> 3) & ~1));
- }
- if (my >= 0)
- my = (h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
- else {
- my = -my;
- my = -(h263_chroma_roundtab[my & 0xf] + ((my >> 3) & ~1));
- }
- dxy = ((my & 1) << 1) | (mx & 1);
- mx >>= 1;
- my >>= 1;
-
- src_x = mb_x * 8 + mx;
- src_y = mb_y * 8 + my;
- src_x = clip(src_x, -8, s->width/2);
- if (src_x == s->width/2)
- dxy &= ~1;
- src_y = clip(src_y, -8, s->height/2);
- if (src_y == s->height/2)
- dxy &= ~2;
-
- offset = (src_y * (s->linesize >> 1)) + src_x;
- ptr = s->last_picture[1] + offset;
- sub_pixels_2(s->block[4], ptr, s->linesize >> 1, dxy);
- ptr = s->last_picture[2] + offset;
- sub_pixels_2(s->block[5], ptr, s->linesize >> 1, dxy);
+ diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
+ diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
}
+
+ /* pre quantization */
+ if(s->mc_mb_var[s->mb_width*mb_y+ mb_x]<2*s->qscale*s->qscale){
+ if(pix_abs8x8(ptr_y , dest_y , wrap_y) < 20*s->qscale) skip_dct[0]= 1;
+ if(pix_abs8x8(ptr_y + 8, dest_y + 8, wrap_y) < 20*s->qscale) skip_dct[1]= 1;
+ if(pix_abs8x8(ptr_y + 8*wrap_y , dest_y + 8*wrap_y , wrap_y) < 20*s->qscale) skip_dct[2]= 1;
+ if(pix_abs8x8(ptr_y + 8*wrap_y + 8, dest_y + 8*wrap_y + 8, wrap_y) < 20*s->qscale) skip_dct[3]= 1;
+ if(pix_abs8x8(ptr_cb , dest_cb , wrap_y) < 20*s->qscale) skip_dct[4]= 1;
+ if(pix_abs8x8(ptr_cr , dest_cr , wrap_y) < 20*s->qscale) skip_dct[5]= 1;
+#if 0
+{
+ static int stat[7];
+ int num=0;
+ for(i=0; i<6; i++)
+ if(skip_dct[i]) num++;
+ stat[num]++;
+
+ if(s->mb_x==0 && s->mb_y==0){
+ for(i=0; i<7; i++){
+ printf("%6d %1d\n", stat[i], i);
+ }
+ }
+}
+#endif
+ }
+
}
#if 0
@@ -1240,17 +1521,47 @@ static void encode_mb(MpegEncContext *s)
}
#endif
/* DCT & quantize */
- if (s->h263_msmpeg4) {
- msmpeg4_dc_scale(s);
- } else if (s->h263_pred) {
+ if (s->h263_pred && !(s->msmpeg4_version==1 || s->msmpeg4_version==2)) {
h263_dc_scale(s);
+ } else if (s->h263_aic) {
+ s->y_dc_scale = 2*s->qscale;
+ s->c_dc_scale = 2*s->qscale;
} else {
/* default quantization values */
s->y_dc_scale = 8;
s->c_dc_scale = 8;
}
- for(i=0;i<6;i++) {
- s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale);
+ if(s->out_format==FMT_MJPEG){
+ for(i=0;i<6;i++) {
+ int overflow;
+ s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow);
+ if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
+ }
+ }else{
+ for(i=0;i<6;i++) {
+ if(!skip_dct[i]){
+ int overflow;
+ s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow);
+ // FIXME we could decide to change to quantizer instead of clipping
+ // JS: I don't think that would be a good idea it could lower quality instead
+ // of improve it. Just INTRADC clipping deserves changes in quantizer
+ if (overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]);
+ }else
+ s->block_last_index[i]= -1;
+ }
+ if(s->luma_elim_threshold && !s->mb_intra)
+ for(i=0; i<4; i++)
+ dct_single_coeff_elimination(s, i, s->luma_elim_threshold, 0);
+ if(s->chroma_elim_threshold && !s->mb_intra)
+ for(i=4; i<6; i++)
+ dct_single_coeff_elimination(s, i, s->chroma_elim_threshold, 1);
+ }
+
+ if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
+ s->block_last_index[4]=
+ s->block_last_index[5]= 0;
+ s->block[4][0]=
+ s->block[5][0]= 128;
}
/* huffman encode */
@@ -1272,14 +1583,113 @@ static void encode_mb(MpegEncContext *s)
}
}
-static void copy_bits(PutBitContext *pb, UINT8 *src, int length)
+void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length)
{
+#if 1
+ int bytes= length>>4;
+ int bits= length&15;
+ int i;
+
+ if(length==0) return;
+
+ for(i=0; i<bytes; i++) put_bits(pb, 16, be2me_16(((uint16_t*)src)[i]));
+ put_bits(pb, bits, be2me_16(((uint16_t*)src)[i])>>(16-bits));
+#else
int bytes= length>>3;
int bits= length&7;
int i;
for(i=0; i<bytes; i++) put_bits(pb, 8, src[i]);
put_bits(pb, bits, src[i]>>(8-bits));
+#endif
+}
+
+static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
+ int i;
+
+ memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
+
+ /* mpeg1 */
+ d->mb_incr= s->mb_incr;
+ for(i=0; i<3; i++)
+ d->last_dc[i]= s->last_dc[i];
+
+ /* statistics */
+ d->mv_bits= s->mv_bits;
+ d->i_tex_bits= s->i_tex_bits;
+ d->p_tex_bits= s->p_tex_bits;
+ d->i_count= s->i_count;
+ d->p_count= s->p_count;
+ d->skip_count= s->skip_count;
+ d->misc_bits= s->misc_bits;
+ d->last_bits= 0;
+
+ d->mb_skiped= s->mb_skiped;
+}
+
+static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
+ int i;
+
+ memcpy(d->mv, s->mv, 2*4*2*sizeof(int));
+ memcpy(d->last_mv, s->last_mv, 2*2*2*sizeof(int)); //FIXME is memcpy faster then a loop?
+
+ /* mpeg1 */
+ d->mb_incr= s->mb_incr;
+ for(i=0; i<3; i++)
+ d->last_dc[i]= s->last_dc[i];
+
+ /* statistics */
+ d->mv_bits= s->mv_bits;
+ d->i_tex_bits= s->i_tex_bits;
+ d->p_tex_bits= s->p_tex_bits;
+ d->i_count= s->i_count;
+ d->p_count= s->p_count;
+ d->skip_count= s->skip_count;
+ d->misc_bits= s->misc_bits;
+
+ d->mb_intra= s->mb_intra;
+ d->mb_skiped= s->mb_skiped;
+ d->mv_type= s->mv_type;
+ d->mv_dir= s->mv_dir;
+ d->pb= s->pb;
+ if(s->data_partitioning){
+ d->pb2= s->pb2;
+ d->tex_pb= s->tex_pb;
+ }
+ d->block= s->block;
+ for(i=0; i<6; i++)
+ d->block_last_index[i]= s->block_last_index[i];
+}
+
+static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegEncContext *best, int type,
+ PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
+ int *dmin, int *next_block, int motion_x, int motion_y)
+{
+ int bits_count;
+
+ copy_context_before_encode(s, backup, type);
+
+ s->block= s->blocks[*next_block];
+ s->pb= pb[*next_block];
+ if(s->data_partitioning){
+ s->pb2 = pb2 [*next_block];
+ s->tex_pb= tex_pb[*next_block];
+ }
+
+ encode_mb(s, motion_x, motion_y);
+
+ bits_count= get_bit_count(&s->pb);
+ if(s->data_partitioning){
+ bits_count+= get_bit_count(&s->pb2);
+ bits_count+= get_bit_count(&s->tex_pb);
+ }
+
+ if(bits_count<*dmin){
+ *dmin= bits_count;
+ *next_block^=1;
+
+ copy_context_after_encode(best, s, type);
+ }
}
static void encode_picture(MpegEncContext *s, int picture_number)
@@ -1287,8 +1697,17 @@ static void encode_picture(MpegEncContext *s, int picture_number)
int mb_x, mb_y, last_gob, pdif = 0;
int i;
int bits;
- MpegEncContext best_s;
- UINT8 bit_buf[4][3000]; //FIXME check that this is ALLWAYS large enogh for a MB
+ MpegEncContext best_s, backup_s;
+ UINT8 bit_buf[2][3000];
+ UINT8 bit_buf2[2][3000];
+ UINT8 bit_buf_tex[2][3000];
+ PutBitContext pb[2], pb2[2], tex_pb[2];
+
+ for(i=0; i<2; i++){
+ init_put_bits(&pb [i], bit_buf [i], 3000, NULL, NULL);
+ init_put_bits(&pb2 [i], bit_buf2 [i], 3000, NULL, NULL);
+ init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL);
+ }
s->picture_number = picture_number;
@@ -1299,12 +1718,16 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->block_wrap[4]=
s->block_wrap[5]= s->mb_width + 2;
- s->last_mc_mb_var = s->mc_mb_var;
/* Reset the average MB variance */
- s->avg_mb_var = 0;
- s->mc_mb_var = 0;
+ s->mb_var_sum = 0;
+ s->mc_mb_var_sum = 0;
+
+ /* we need to initialize some time vars before we can encode b-frames */
+ if (s->h263_pred && !s->h263_msmpeg4)
+ ff_set_mpeg4_time(s, s->picture_number);
+
/* Estimate motion for every MB */
- if(s->pict_type == P_TYPE){
+ if(s->pict_type != I_TYPE){
for(mb_y=0; mb_y < s->mb_height; mb_y++) {
s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
@@ -1319,127 +1742,61 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->block_index[3]+=2;
/* compute motion vector & mb_type and store in context */
- estimate_motion(s, mb_x, mb_y);
+ if(s->pict_type==B_TYPE)
+ ff_estimate_b_frame_motion(s, mb_x, mb_y);
+ else
+ ff_estimate_p_frame_motion(s, mb_x, mb_y);
// s->mb_type[mb_y*s->mb_width + mb_x]=MB_TYPE_INTER;
}
}
emms_c();
- }else{
+ }else /* if(s->pict_type == I_TYPE) */{
/* I-Frame */
//FIXME do we need to zero them?
memset(s->motion_val[0], 0, sizeof(INT16)*(s->mb_width*2 + 2)*(s->mb_height*2 + 2)*2);
- memset(s->mv_table[0] , 0, sizeof(INT16)*s->mb_width*s->mb_height);
- memset(s->mv_table[1] , 0, sizeof(INT16)*s->mb_width*s->mb_height);
+ memset(s->p_mv_table , 0, sizeof(INT16)*(s->mb_width+2)*(s->mb_height+2)*2);
memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
}
- if(s->avg_mb_var < s->mc_mb_var && s->pict_type != B_TYPE && (!s->force_type)){ //FIXME subtract MV bits
+ if(s->mb_var_sum < s->mc_mb_var_sum && s->pict_type == P_TYPE){ //FIXME subtract MV bits
s->pict_type= I_TYPE;
- s->picture_in_gop_number=0;
memset(s->mb_type , MB_TYPE_INTRA, sizeof(UINT8)*s->mb_width*s->mb_height);
+ if(s->max_b_frames==0){
+ s->input_pict_type= I_TYPE;
+ s->input_picture_in_gop_number=0;
+ }
//printf("Scene change detected, encoding as I Frame\n");
}
-
- /* find best f_code for ME which do unlimited searches */
- if(s->pict_type==P_TYPE && s->full_search>3){
- int mv_num[8];
- int i;
- int loose=0;
- UINT8 * fcode_tab= s->fcode_tab;
-
- for(i=0; i<8; i++) mv_num[i]=0;
-
- for(i=0; i<s->mb_num; i++){
- if(s->mb_type[i] & MB_TYPE_INTER){
- mv_num[ fcode_tab[s->mv_table[0][i] + MAX_MV] ]++;
- mv_num[ fcode_tab[s->mv_table[1][i] + MAX_MV] ]++;
-//printf("%d %d %d\n", s->mv_table[0][i], fcode_tab[s->mv_table[0][i] + MAX_MV], i);
- }
-//else printf("I");
- }
-
- for(i=MAX_FCODE; i>1; i--){
- loose+= mv_num[i];
- if(loose > 10) break; //FIXME this is pretty ineffective
- }
- s->f_code= i;
-/* for(i=0; i<=MAX_FCODE; i++){
- printf("%d ", mv_num[i]);
- }
- printf("\n");*/
- }else{
- s->f_code= 1;
+
+ if(s->pict_type==P_TYPE || s->pict_type==S_TYPE)
+ s->f_code= ff_get_best_fcode(s, s->p_mv_table, MB_TYPE_INTER);
+ ff_fix_long_p_mvs(s);
+ if(s->pict_type==B_TYPE){
+ s->f_code= ff_get_best_fcode(s, s->b_forw_mv_table, MB_TYPE_FORWARD);
+ s->b_code= ff_get_best_fcode(s, s->b_back_mv_table, MB_TYPE_BACKWARD);
+
+ ff_fix_long_b_mvs(s, s->b_forw_mv_table, s->f_code, MB_TYPE_FORWARD);
+ ff_fix_long_b_mvs(s, s->b_back_mv_table, s->b_code, MB_TYPE_BACKWARD);
+ ff_fix_long_b_mvs(s, s->b_bidir_forw_mv_table, s->f_code, MB_TYPE_BIDIR);
+ ff_fix_long_b_mvs(s, s->b_bidir_back_mv_table, s->b_code, MB_TYPE_BIDIR);
}
-
+
//printf("f_code %d ///\n", s->f_code);
- /* convert MBs with too long MVs to I-Blocks */
- if(s->pict_type==P_TYPE){
- int i, x, y;
- const int f_code= s->f_code;
- UINT8 * fcode_tab= s->fcode_tab;
-//FIXME try to clip instead of intra izing ;)
- /* clip / convert to intra 16x16 type MVs */
- for(i=0; i<s->mb_num; i++){
- if(s->mb_type[i]&MB_TYPE_INTER){
- if( fcode_tab[s->mv_table[0][i] + MAX_MV] > f_code
- || fcode_tab[s->mv_table[0][i] + MAX_MV] == 0
- || fcode_tab[s->mv_table[1][i] + MAX_MV] > f_code
- || fcode_tab[s->mv_table[1][i] + MAX_MV] == 0 ){
- s->mb_type[i] &= ~MB_TYPE_INTER;
- s->mb_type[i] |= MB_TYPE_INTRA;
- s->mv_table[0][i] = 0;
- s->mv_table[1][i] = 0;
- }
- }
- }
-
- if(s->flags&CODEC_FLAG_4MV){
- int wrap= 2+ s->mb_width*2;
-
- /* clip / convert to intra 8x8 type MVs */
- for(y=0; y<s->mb_height; y++){
- int xy= (y*2 + 1)*wrap + 1;
- i= y*s->mb_width;
-
- for(x=0; x<s->mb_width; x++){
- if(s->mb_type[i]&MB_TYPE_INTER4V){
- int block;
- for(block=0; block<4; block++){
- int off= (block& 1) + (block>>1)*wrap;
- int mx= s->motion_val[ xy + off ][0];
- int my= s->motion_val[ xy + off ][1];
-
- if( fcode_tab[mx + MAX_MV] > f_code
- || fcode_tab[mx + MAX_MV] == 0
- || fcode_tab[my + MAX_MV] > f_code
- || fcode_tab[my + MAX_MV] == 0 ){
- s->mb_type[i] &= ~MB_TYPE_INTER4V;
- s->mb_type[i] |= MB_TYPE_INTRA;
- }
- }
- xy+=2;
- i++;
- }
- }
- }
- }
- }
// printf("%d %d\n", s->avg_mb_var, s->mc_mb_var);
- if (!s->fixed_qscale)
- s->qscale = rate_estimate_qscale(s);
+ if(s->flags&CODEC_FLAG_PASS2)
+ s->qscale = ff_rate_estimate_qscale_pass2(s);
+ else if (!s->fixed_qscale)
+ s->qscale = ff_rate_estimate_qscale(s);
- /* precompute matrix */
if (s->out_format == FMT_MJPEG) {
/* for mjpeg, we do include qscale in the matrix */
s->intra_matrix[0] = default_intra_matrix[0];
for(i=1;i<64;i++)
- s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3;
- convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8);
- } else {
- convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale);
- convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale);
+ s->intra_matrix[i] = CLAMP_TO_8BIT((default_intra_matrix[i] * s->qscale) >> 3);
+ convert_matrix(s->q_intra_matrix, s->q_intra_matrix16,
+ s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias);
}
s->last_bits= get_bit_count(&s->pb);
@@ -1489,21 +1846,31 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->gob_index = 2;
else
s->gob_index = 4;
+ }else if(s->codec_id==CODEC_ID_MPEG4){
+ s->gob_index = 1;
}
-
- s->avg_mb_var = s->avg_mb_var / s->mb_num;
-
+
+ if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
+ ff_mpeg4_init_partitions(s);
+
+ s->resync_mb_x=0;
+ s->resync_mb_y=0;
for(mb_y=0; mb_y < s->mb_height; mb_y++) {
- /* Put GOB header based on RTP MTU */
+ /* Put GOB header based on RTP MTU for formats which support it per line (H263*)*/
/* TODO: Put all this stuff in a separate generic function */
if (s->rtp_mode) {
if (!mb_y) {
s->ptr_lastgob = s->pb.buf;
s->ptr_last_mb_line = s->pb.buf;
} else if (s->out_format == FMT_H263 && !s->h263_pred && !s->h263_msmpeg4 && !(mb_y % s->gob_index)) {
+ // MN: we could move the space check from h263 -> here, as its not h263 specific
last_gob = h263_encode_gob_header(s, mb_y);
if (last_gob) {
- s->first_gob_line = 1;
+ s->first_slice_line = 1;
+ }else{
+ /*MN: we reset it here instead at the end of each line cuz mpeg4 can have
+ slice lines starting & ending in the middle*/
+ s->first_slice_line = 0;
}
}
}
@@ -1516,10 +1883,9 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
for(mb_x=0; mb_x < s->mb_width; mb_x++) {
const int mb_type= s->mb_type[mb_y * s->mb_width + mb_x];
- PutBitContext pb;
- int d;
+ const int xy= (mb_y+1) * (s->mb_width+2) + mb_x + 1;
+// int d;
int dmin=10000000;
- int best=0;
s->mb_x = mb_x;
s->mb_y = mb_y;
@@ -1529,114 +1895,216 @@ static void encode_picture(MpegEncContext *s, int picture_number)
s->block_index[3]+=2;
s->block_index[4]++;
s->block_index[5]++;
+
+ /* write gob / video packet header for formats which support it at any MB (MPEG4) */
+ if(s->rtp_mode && s->mb_y>0 && s->codec_id==CODEC_ID_MPEG4){
+ int pdif= pbBufPtr(&s->pb) - s->ptr_lastgob;
+
+ //the *2 is there so we stay below the requested size
+ if(pdif + s->mb_line_avgsize/s->mb_width >= s->rtp_payload_size){
+ if(s->codec_id==CODEC_ID_MPEG4){
+ if(s->data_partitioning && s->pict_type!=B_TYPE){
+ ff_mpeg4_merge_partitions(s);
+ ff_mpeg4_init_partitions(s);
+ }
+ ff_mpeg4_encode_video_packet_header(s);
+
+ if(s->flags&CODEC_FLAG_PASS1){
+ int bits= get_bit_count(&s->pb);
+ s->misc_bits+= bits - s->last_bits;
+ s->last_bits= bits;
+ }
+ ff_mpeg4_clean_buffers(s);
+ }
+ s->ptr_lastgob = pbBufPtr(&s->pb);
+ s->first_slice_line=1;
+ s->resync_mb_x=mb_x;
+ s->resync_mb_y=mb_y;
+ }
+
+ if( (s->resync_mb_x == s->mb_x)
+ && s->resync_mb_y+1 == s->mb_y){
+ s->first_slice_line=0;
+ }
+ }
- s->mv_dir = MV_DIR_FORWARD;
if(mb_type & (mb_type-1)){ // more than 1 MB type possible
- pb= s->pb;
+ int next_block=0;
+ int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
+
+ copy_context_before_encode(&backup_s, s, -1);
+ backup_s.pb= s->pb;
+ best_s.data_partitioning= s->data_partitioning;
+ if(s->data_partitioning){
+ backup_s.pb2= s->pb2;
+ backup_s.tex_pb= s->tex_pb;
+ }
+
if(mb_type&MB_TYPE_INTER){
+ s->mv_dir = MV_DIR_FORWARD;
s->mv_type = MV_TYPE_16X16;
s->mb_intra= 0;
- s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x];
- s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x];
- init_put_bits(&s->pb, bit_buf[1], 3000, NULL, NULL);
- s->block= s->inter_block;
-
- encode_mb(s);
- d= get_bit_count(&s->pb);
- if(d<dmin){
- flush_put_bits(&s->pb);
- dmin=d;
- best_s.mv[0][0][0]= s->mv[0][0][0];
- best_s.mv[0][0][1]= s->mv[0][0][1];
- best_s.mb_intra= 0;
- best_s.mv_type = MV_TYPE_16X16;
- best_s.pb=s->pb;
- best_s.block= s->block;
- best=1;
- for(i=0; i<6; i++)
- best_s.block_last_index[i]= s->block_last_index[i];
- }
+ s->mv[0][0][0] = s->p_mv_table[xy][0];
+ s->mv[0][0][1] = s->p_mv_table[xy][1];
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb,
+ &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
}
- if(mb_type&MB_TYPE_INTER4V){
+ if(mb_type&MB_TYPE_INTER4V){
+ s->mv_dir = MV_DIR_FORWARD;
s->mv_type = MV_TYPE_8X8;
s->mb_intra= 0;
for(i=0; i<4; i++){
s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
}
- init_put_bits(&s->pb, bit_buf[2], 3000, NULL, NULL);
- s->block= s->inter4v_block;
-
- encode_mb(s);
- d= get_bit_count(&s->pb);
- if(d<dmin){
- flush_put_bits(&s->pb);
- dmin=d;
- for(i=0; i<4; i++){
- best_s.mv[0][i][0] = s->mv[0][i][0];
- best_s.mv[0][i][1] = s->mv[0][i][1];
- }
- best_s.mb_intra= 0;
- best_s.mv_type = MV_TYPE_8X8;
- best_s.pb=s->pb;
- best_s.block= s->block;
- best=2;
- for(i=0; i<6; i++)
- best_s.block_last_index[i]= s->block_last_index[i];
- }
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER4V, pb, pb2, tex_pb,
+ &dmin, &next_block, 0, 0);
+ }
+ if(mb_type&MB_TYPE_FORWARD){
+ s->mv_dir = MV_DIR_FORWARD;
+ s->mv_type = MV_TYPE_16X16;
+ s->mb_intra= 0;
+ s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
+ s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_FORWARD, pb, pb2, tex_pb,
+ &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
+ }
+ if(mb_type&MB_TYPE_BACKWARD){
+ s->mv_dir = MV_DIR_BACKWARD;
+ s->mv_type = MV_TYPE_16X16;
+ s->mb_intra= 0;
+ s->mv[1][0][0] = s->b_back_mv_table[xy][0];
+ s->mv[1][0][1] = s->b_back_mv_table[xy][1];
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BACKWARD, pb, pb2, tex_pb,
+ &dmin, &next_block, s->mv[1][0][0], s->mv[1][0][1]);
+ }
+ if(mb_type&MB_TYPE_BIDIR){
+ s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
+ s->mv_type = MV_TYPE_16X16;
+ s->mb_intra= 0;
+ s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
+ s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
+ s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
+ s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_BIDIR, pb, pb2, tex_pb,
+ &dmin, &next_block, 0, 0);
+ }
+ if(mb_type&MB_TYPE_DIRECT){
+ s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
+ s->mv_type = MV_TYPE_16X16; //FIXME
+ s->mb_intra= 0;
+ s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
+ s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
+ s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
+ s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_DIRECT, pb, pb2, tex_pb,
+ &dmin, &next_block, s->b_direct_mv_table[xy][0], s->b_direct_mv_table[xy][1]);
}
if(mb_type&MB_TYPE_INTRA){
+ s->mv_dir = MV_DIR_FORWARD;
s->mv_type = MV_TYPE_16X16;
s->mb_intra= 1;
s->mv[0][0][0] = 0;
s->mv[0][0][1] = 0;
- init_put_bits(&s->pb, bit_buf[0], 3000, NULL, NULL);
- s->block= s->intra_block;
-
- encode_mb(s);
- d= get_bit_count(&s->pb);
- if(d<dmin){
- flush_put_bits(&s->pb);
- dmin=d;
- best_s.mv[0][0][0]= 0;
- best_s.mv[0][0][1]= 0;
- best_s.mb_intra= 1;
- best_s.mv_type = MV_TYPE_16X16;
- best_s.pb=s->pb;
- best_s.block= s->block;
- for(i=0; i<6; i++)
- best_s.block_last_index[i]= s->block_last_index[i];
- best=0;
- }
- /* force cleaning of ac/dc if needed ... */
- s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
+ encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb,
+ &dmin, &next_block, 0, 0);
+ /* force cleaning of ac/dc pred stuff if needed ... */
+ if(s->h263_pred || s->h263_aic)
+ s->mbintra_table[mb_x + mb_y*s->mb_width]=1;
}
- for(i=0; i<4; i++){
- s->mv[0][i][0] = best_s.mv[0][i][0];
- s->mv[0][i][1] = best_s.mv[0][i][1];
+ copy_context_after_encode(s, &best_s, -1);
+
+ pb_bits_count= get_bit_count(&s->pb);
+ flush_put_bits(&s->pb);
+ ff_copy_bits(&backup_s.pb, bit_buf[next_block^1], pb_bits_count);
+ s->pb= backup_s.pb;
+
+ if(s->data_partitioning){
+ pb2_bits_count= get_bit_count(&s->pb2);
+ flush_put_bits(&s->pb2);
+ ff_copy_bits(&backup_s.pb2, bit_buf2[next_block^1], pb2_bits_count);
+ s->pb2= backup_s.pb2;
+
+ tex_pb_bits_count= get_bit_count(&s->tex_pb);
+ flush_put_bits(&s->tex_pb);
+ ff_copy_bits(&backup_s.tex_pb, bit_buf_tex[next_block^1], tex_pb_bits_count);
+ s->tex_pb= backup_s.tex_pb;
}
- s->mb_intra= best_s.mb_intra;
- s->mv_type= best_s.mv_type;
- for(i=0; i<6; i++)
- s->block_last_index[i]= best_s.block_last_index[i];
- copy_bits(&pb, bit_buf[best], dmin);
- s->block= best_s.block;
- s->pb= pb;
+ s->last_bits= get_bit_count(&s->pb);
} else {
+ int motion_x, motion_y;
+ s->mv_type=MV_TYPE_16X16;
// only one MB-Type possible
- if(mb_type&MB_TYPE_INTRA){
+ switch(mb_type){
+ case MB_TYPE_INTRA:
+ s->mv_dir = MV_DIR_FORWARD;
s->mb_intra= 1;
- s->mv[0][0][0] = 0;
- s->mv[0][0][1] = 0;
- }else{
+ motion_x= s->mv[0][0][0] = 0;
+ motion_y= s->mv[0][0][1] = 0;
+ break;
+ case MB_TYPE_INTER:
+ s->mv_dir = MV_DIR_FORWARD;
+ s->mb_intra= 0;
+ motion_x= s->mv[0][0][0] = s->p_mv_table[xy][0];
+ motion_y= s->mv[0][0][1] = s->p_mv_table[xy][1];
+ break;
+ case MB_TYPE_INTER4V:
+ s->mv_dir = MV_DIR_FORWARD;
+ s->mv_type = MV_TYPE_8X8;
s->mb_intra= 0;
- s->mv[0][0][0] = s->mv_table[0][mb_y * s->mb_width + mb_x];
- s->mv[0][0][1] = s->mv_table[1][mb_y * s->mb_width + mb_x];
+ for(i=0; i<4; i++){
+ s->mv[0][i][0] = s->motion_val[s->block_index[i]][0];
+ s->mv[0][i][1] = s->motion_val[s->block_index[i]][1];
+ }
+ motion_x= motion_y= 0;
+ break;
+ case MB_TYPE_DIRECT:
+ s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD | MV_DIRECT;
+ s->mb_intra= 0;
+ motion_x=s->b_direct_mv_table[xy][0];
+ motion_y=s->b_direct_mv_table[xy][1];
+ s->mv[0][0][0] = s->b_direct_forw_mv_table[xy][0];
+ s->mv[0][0][1] = s->b_direct_forw_mv_table[xy][1];
+ s->mv[1][0][0] = s->b_direct_back_mv_table[xy][0];
+ s->mv[1][0][1] = s->b_direct_back_mv_table[xy][1];
+ break;
+ case MB_TYPE_BIDIR:
+ s->mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
+ s->mb_intra= 0;
+ motion_x=0;
+ motion_y=0;
+ s->mv[0][0][0] = s->b_bidir_forw_mv_table[xy][0];
+ s->mv[0][0][1] = s->b_bidir_forw_mv_table[xy][1];
+ s->mv[1][0][0] = s->b_bidir_back_mv_table[xy][0];
+ s->mv[1][0][1] = s->b_bidir_back_mv_table[xy][1];
+ break;
+ case MB_TYPE_BACKWARD:
+ s->mv_dir = MV_DIR_BACKWARD;
+ s->mb_intra= 0;
+ motion_x= s->mv[1][0][0] = s->b_back_mv_table[xy][0];
+ motion_y= s->mv[1][0][1] = s->b_back_mv_table[xy][1];
+ break;
+ case MB_TYPE_FORWARD:
+ s->mv_dir = MV_DIR_FORWARD;
+ s->mb_intra= 0;
+ motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
+ motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
+// printf(" %d %d ", motion_x, motion_y);
+ break;
+ default:
+ motion_x=motion_y=0; //gcc warning fix
+ printf("illegal MB type\n");
}
- encode_mb(s);
+ encode_mb(s, motion_x, motion_y);
+ }
+ /* clean the MV table in IPS frames for direct mode in B frames */
+ if(s->mb_intra /* && I,P,S_TYPE */){
+ s->p_mv_table[xy][0]=0;
+ s->p_mv_table[xy][1]=0;
}
MPV_decode_mb(s, s->block);
+//printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_width, get_bit_count(&s->pb));
}
@@ -1650,14 +2118,20 @@ static void encode_picture(MpegEncContext *s, int picture_number)
}
//fprintf(stderr, "\nMB line: %d\tSize: %u\tAvg. Size: %u", s->mb_y,
// (s->pb.buf_ptr - s->ptr_last_mb_line), s->mb_line_avgsize);
- s->first_gob_line = 0;
+ if(s->codec_id!=CODEC_ID_MPEG4) s->first_slice_line = 0; //FIXME clean
}
}
emms_c();
- if (s->h263_msmpeg4 && s->pict_type == I_TYPE)
+ if(s->codec_id==CODEC_ID_MPEG4 && s->data_partitioning && s->pict_type!=B_TYPE)
+ ff_mpeg4_merge_partitions(s);
+
+ if (s->msmpeg4_version && s->msmpeg4_version<4 && s->pict_type == I_TYPE)
msmpeg4_encode_ext_header(s);
+ if(s->codec_id==CODEC_ID_MPEG4)
+ ff_mpeg4_stuffing(&s->pb);
+
//if (s->gob_number)
// fprintf(stderr,"\nNumber of GOB: %d", s->gob_number);
@@ -1675,30 +2149,14 @@ static void encode_picture(MpegEncContext *s, int picture_number)
static int dct_quantize_c(MpegEncContext *s,
DCTELEM *block, int n,
- int qscale)
+ int qscale, int *overflow)
{
int i, j, level, last_non_zero, q;
const int *qmat;
- int minLevel, maxLevel;
-
- if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){
- /* mpeg4 */
- minLevel= -2048;
- maxLevel= 2047;
- }else if(s->out_format==FMT_MPEG1){
- /* mpeg1 */
- minLevel= -255;
- maxLevel= 255;
- }else if(s->out_format==FMT_MJPEG){
- /* (m)jpeg */
- minLevel= -1023;
- maxLevel= 1023;
- }else{
- /* h263 / msmpeg4 */
- minLevel= -128;
- maxLevel= 127;
- }
-
+ int bias;
+ int max=0;
+ unsigned int threshold1, threshold2;
+
av_fdct (block);
/* we need this permutation so that we correct the IDCT
@@ -1706,81 +2164,54 @@ static int dct_quantize_c(MpegEncContext *s,
block_permute(block);
if (s->mb_intra) {
- if (n < 4)
- q = s->y_dc_scale;
- else
- q = s->c_dc_scale;
- q = q << 3;
-
+ if (!s->h263_aic) {
+ if (n < 4)
+ q = s->y_dc_scale;
+ else
+ q = s->c_dc_scale;
+ q = q << 3;
+ } else
+ /* For AIC we skip quant/dequant of INTRADC */
+ q = 1 << 3;
+
/* note: block[0] is assumed to be positive */
block[0] = (block[0] + (q >> 1)) / q;
i = 1;
last_non_zero = 0;
- if (s->out_format == FMT_H263) {
- qmat = s->q_non_intra_matrix;
- } else {
- qmat = s->q_intra_matrix;
- }
+ qmat = s->q_intra_matrix[qscale];
+ bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
} else {
i = 0;
last_non_zero = -1;
- qmat = s->q_non_intra_matrix;
+ qmat = s->q_inter_matrix[qscale];
+ bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT);
}
+ threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1;
+ threshold2= threshold1<<1;
for(;i<64;i++) {
j = zigzag_direct[i];
level = block[j];
level = level * qmat[j];
-#ifdef PARANOID
- {
- static int count = 0;
- int level1, level2, qmat1;
- double val;
- if (qmat == s->q_non_intra_matrix) {
- qmat1 = default_non_intra_matrix[j] * s->qscale;
- } else {
- qmat1 = default_intra_matrix[j] * s->qscale;
- }
- if (av_fdct != jpeg_fdct_ifast)
- val = ((double)block[j] * 8.0) / (double)qmat1;
- else
- val = ((double)block[j] * 8.0 * 2048.0) /
- ((double)qmat1 * aanscales[j]);
- level1 = (int)val;
- level2 = level / (1 << (QMAT_SHIFT - 3));
- if (level1 != level2) {
- fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n",
- count, level2, level1, block[j], qmat1, qmat[j],
- val);
- count++;
- }
- }
-#endif
- /* XXX: slight error for the low range. Test should be equivalent to
- (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 <<
- (QMAT_SHIFT - 3)))
- */
- if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) !=
- level) {
- level = level / (1 << (QMAT_SHIFT - 3));
- /* XXX: currently, this code is not optimal. the range should be:
- mpeg1: -255..255
- mpeg2: -2048..2047
- h263: -128..127
- mpeg4: -2048..2047
- */
- if (level > maxLevel)
- level = maxLevel;
- else if (level < minLevel)
- level = minLevel;
-
- block[j] = level;
+// if( bias+level >= (1<<(QMAT_SHIFT - 3))
+// || bias-level >= (1<<(QMAT_SHIFT - 3))){
+ if(((unsigned)(level+threshold1))>threshold2){
+ if(level>0){
+ level= (bias + level)>>(QMAT_SHIFT - 3);
+ block[j]= level;
+ }else{
+ level= (bias - level)>>(QMAT_SHIFT - 3);
+ block[j]= -level;
+ }
+ max |=level;
last_non_zero = i;
- } else {
- block[j] = 0;
+ }else{
+ block[j]=0;
}
}
+ *overflow= s->max_qcoeff < max; //overflow might have happend
+
return last_non_zero;
}
@@ -1822,7 +2253,7 @@ static void dct_unquantize_mpeg1_c(MpegEncContext *s,
}
} else {
i = 0;
- quant_matrix = s->non_intra_matrix;
+ quant_matrix = s->inter_matrix;
for(;i<nCoeffs;i++) {
int j= zigzag_direct[i];
level = block[j];
@@ -1848,6 +2279,69 @@ static void dct_unquantize_mpeg1_c(MpegEncContext *s,
}
}
+static void dct_unquantize_mpeg2_c(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int i, level, nCoeffs;
+ const UINT16 *quant_matrix;
+
+ if(s->alternate_scan) nCoeffs= 64;
+ else nCoeffs= s->block_last_index[n]+1;
+
+ if (s->mb_intra) {
+ if (n < 4)
+ block[0] = block[0] * s->y_dc_scale;
+ else
+ block[0] = block[0] * s->c_dc_scale;
+ quant_matrix = s->intra_matrix;
+ for(i=1;i<nCoeffs;i++) {
+ int j= zigzag_direct[i];
+ level = block[j];
+ if (level) {
+ if (level < 0) {
+ level = -level;
+ level = (int)(level * qscale * quant_matrix[j]) >> 3;
+ level = -level;
+ } else {
+ level = (int)(level * qscale * quant_matrix[j]) >> 3;
+ }
+#ifdef PARANOID
+ if (level < -2048 || level > 2047)
+ fprintf(stderr, "unquant error %d %d\n", i, level);
+#endif
+ block[j] = level;
+ }
+ }
+ } else {
+ int sum=-1;
+ i = 0;
+ quant_matrix = s->inter_matrix;
+ for(;i<nCoeffs;i++) {
+ int j= zigzag_direct[i];
+ level = block[j];
+ if (level) {
+ if (level < 0) {
+ level = -level;
+ level = (((level << 1) + 1) * qscale *
+ ((int) (quant_matrix[j]))) >> 4;
+ level = -level;
+ } else {
+ level = (((level << 1) + 1) * qscale *
+ ((int) (quant_matrix[j]))) >> 4;
+ }
+#ifdef PARANOID
+ if (level < -2048 || level > 2047)
+ fprintf(stderr, "unquant error %d %d\n", i, level);
+#endif
+ block[j] = level;
+ sum+=level;
+ }
+ }
+ block[63]^=sum&1;
+ }
+}
+
+
static void dct_unquantize_h263_c(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
@@ -1891,188 +2385,117 @@ static void dct_unquantize_h263_c(MpegEncContext *s,
}
}
-/* rate control */
-
-/* an I frame is I_FRAME_SIZE_RATIO bigger than a P frame */
-#define I_FRAME_SIZE_RATIO 3.0
-#define QSCALE_K 20
-
-static void rate_control_init(MpegEncContext *s)
+static void remove_ac(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y)
{
-#if 1
- emms_c();
-
- //initial values, they dont really matter as they will be totally different within a few frames
- s->i_pred.coeff= s->p_pred.coeff= 7.0;
- s->i_pred.count= s->p_pred.count= 1.0;
-
- s->i_pred.decay= s->p_pred.decay= 0.4;
-
- // use more bits at the beginning, otherwise high motion at the begin will look like shit
- s->qsum=100;
- s->qcount=100;
-
- s->short_term_qsum=0.001;
- s->short_term_qcount=0.001;
-#else
- s->wanted_bits = 0;
-
- if (s->intra_only) {
- s->I_frame_bits = ((INT64)s->bit_rate * FRAME_RATE_BASE) / s->frame_rate;
- s->P_frame_bits = s->I_frame_bits;
- } else {
- s->P_frame_bits = (int) ((float)(s->gop_size * s->bit_rate) /
- (float)((float)s->frame_rate / FRAME_RATE_BASE * (I_FRAME_SIZE_RATIO + s->gop_size - 1)));
- s->I_frame_bits = (int)(s->P_frame_bits * I_FRAME_SIZE_RATIO);
+ int dc, dcb, dcr, y, i;
+ for(i=0; i<4; i++){
+ dc= s->dc_val[0][mb_x*2+1 + (i&1) + (mb_y*2+1 + (i>>1))*(s->mb_width*2+2)];
+ for(y=0; y<8; y++){
+ int x;
+ for(x=0; x<8; x++){
+ dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8;
+ }
+ }
+ }
+ dcb = s->dc_val[1][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
+ dcr= s->dc_val[2][mb_x+1 + (mb_y+1)*(s->mb_width+2)];
+ for(y=0; y<8; y++){
+ int x;
+ for(x=0; x<8; x++){
+ dest_cb[x + y*(s->linesize>>1)]= dcb/8;
+ dest_cr[x + y*(s->linesize>>1)]= dcr/8;
+ }
}
-
-#if defined(DEBUG)
- printf("I_frame_size=%d P_frame_size=%d\n",
- s->I_frame_bits, s->P_frame_bits);
-#endif
-#endif
-}
-
-static double predict(Predictor *p, double q, double var)
-{
- return p->coeff*var / (q*p->count);
-}
-
-static void update_predictor(Predictor *p, double q, double var, double size)
-{
- double new_coeff= size*q / (var + 1);
- if(var<1000) return;
-/*{
-int pred= predict(p, q, var);
-int error= abs(pred-size);
-static double sum=0;
-static int count=0;
-if(count>5) sum+=error;
-count++;
-if(256*256*256*64%count==0){
- printf("%d %f %f\n", count, sum/count, p->coeff);
-}
-}*/
- p->count*= p->decay;
- p->coeff*= p->decay;
- p->count++;
- p->coeff+= new_coeff;
}
-static int rate_estimate_qscale(MpegEncContext *s)
+/**
+ * will conceal past errors, and allso drop b frames if needed
+ *
+ */
+void ff_conceal_past_errors(MpegEncContext *s, int unknown_pos)
{
-#if 1
- int qmin= s->qmin;
- int qmax= s->qmax;
- int rate_q=5;
- float q;
- int qscale;
- float br_compensation;
- double diff;
- double short_term_q;
- double long_term_q;
- int last_qscale= s->qscale;
- double fps;
- INT64 wanted_bits;
- emms_c();
-
- fps= (double)s->frame_rate / FRAME_RATE_BASE;
- wanted_bits= s->bit_rate*(double)s->picture_number/fps;
-
+ int mb_x= s->mb_x;
+ int mb_y= s->mb_y;
+ int mb_dist=0;
+ int i, intra_count=0, inter_count=0;
+ int intra_conceal= s->msmpeg4_version ? 50 : 50; //FIXME finetune
+ int inter_conceal= s->msmpeg4_version ? 50 : 50;
- if(s->picture_number>2){
- /* update predictors */
- if(s->last_pict_type == I_TYPE){
- //FIXME
- }else{ //P Frame
-//printf("%d %d %d %f\n", s->qscale, s->last_mc_mb_var, s->frame_bits, s->p_pred.coeff);
- update_predictor(&s->p_pred, s->qscale, s->last_mc_mb_var, s->frame_bits);
- }
- }
+ // for last block
+ if(mb_x>=s->mb_width) mb_x= s->mb_width -1;
+ if(mb_y>=s->mb_height) mb_y= s->mb_height-1;
- if(s->pict_type == I_TYPE){
- //FIXME
- rate_q= s->qsum/s->qcount;
- }else{ //P Frame
- int i;
- int diff, best_diff=1000000000;
- for(i=1; i<=31; i++){
- diff= predict(&s->p_pred, i, s->mc_mb_var) - (double)s->bit_rate/fps;
- if(diff<0) diff= -diff;
- if(diff<best_diff){
- best_diff= diff;
- rate_q= i;
- }
- }
+ if(s->decoding_error==0 && unknown_pos){
+ if(s->data_partitioning && s->pict_type!=B_TYPE)
+ s->decoding_error= DECODING_AC_LOST;
+ else
+ s->decoding_error= DECODING_DESYNC;
}
- s->short_term_qsum*=s->qblur;
- s->short_term_qcount*=s->qblur;
+ if(s->decoding_error==DECODING_DESYNC && s->pict_type!=B_TYPE) s->next_p_frame_damaged=1;
- s->short_term_qsum+= rate_q;
- s->short_term_qcount++;
- short_term_q= s->short_term_qsum/s->short_term_qcount;
+ for(i=mb_x + mb_y*s->mb_width; i>=0; i--){
+ if(s->mbintra_table[i]) intra_count++;
+ else inter_count++;
+ }
- long_term_q= s->qsum/s->qcount*(s->total_bits+1)/(wanted_bits+1); //+1 to avoid nan & 0
-
-// q= (long_term_q - short_term_q)*s->qcompress + short_term_q;
- q= 1/((1/long_term_q - 1/short_term_q)*s->qcompress + 1/short_term_q);
+ if(s->decoding_error==DECODING_AC_LOST){
+ intra_conceal*=2;
+ inter_conceal*=2;
+ }else if(s->decoding_error==DECODING_ACDC_LOST){
+ intra_conceal*=2;
+ inter_conceal*=2;
+ }
- diff= s->total_bits - wanted_bits;
- br_compensation= (s->bit_rate_tolerance - diff)/s->bit_rate_tolerance;
- if(br_compensation<=0.0) br_compensation=0.001;
- q/=br_compensation;
+ if(unknown_pos && (intra_count<inter_count)){
+ intra_conceal= inter_conceal= s->mb_num;
+// printf("%d %d\n",intra_count, inter_count);
+ }
- qscale= (int)(q + 0.5);
- if (qscale<qmin) qscale=qmin;
- else if(qscale>qmax) qscale=qmax;
-
- if (qscale<last_qscale-s->max_qdiff) qscale=last_qscale-s->max_qdiff;
- else if(qscale>last_qscale+s->max_qdiff) qscale=last_qscale+s->max_qdiff;
+ fprintf(stderr, "concealing errors\n");
+
+ /* for all MBs from the current one back until the last resync marker */
+ for(; mb_y>=0 && mb_y>=s->resync_mb_y; mb_y--){
+ for(; mb_x>=0; mb_x--){
+ uint8_t *dest_y = s->current_picture[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
+ uint8_t *dest_cb = s->current_picture[1] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
+ uint8_t *dest_cr = s->current_picture[2] + (mb_y * 8 * (s->linesize >> 1)) + mb_x * 8;
+ int mb_x_backup= s->mb_x; //FIXME pass xy to mpeg_motion
+ int mb_y_backup= s->mb_y;
+ s->mb_x=mb_x;
+ s->mb_y=mb_y;
+ if(s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<intra_conceal){
+ if(s->decoding_error==DECODING_AC_LOST){
+ remove_ac(s, dest_y, dest_cb, dest_cr, mb_x, mb_y);
+// printf("remove ac to %d %d\n", mb_x, mb_y);
+ }else{
+ mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
+ s->last_picture, 0, 0, put_pixels_tab,
+ 0/*mx*/, 0/*my*/, 16);
+ }
+ }
+ else if(!s->mbintra_table[mb_y*s->mb_width + mb_x] && mb_dist<inter_conceal){
+ int mx=0;
+ int my=0;
+
+ if(s->decoding_error!=DECODING_DESYNC){
+ int xy= mb_x*2+1 + (mb_y*2+1)*(s->mb_width*2+2);
+ mx= s->motion_val[ xy ][0];
+ my= s->motion_val[ xy ][1];
+ }
- s->qsum+= qscale;
- s->qcount++;
+ mpeg_motion(s, dest_y, dest_cb, dest_cr, 0,
+ s->last_picture, 0, 0, put_pixels_tab,
+ mx, my, 16);
+ }
+ s->mb_x= mb_x_backup;
+ s->mb_y= mb_y_backup;
- s->last_pict_type= s->pict_type;
-//printf("q:%d diff:%d comp:%f rate_q:%d st_q:%f fvar:%d last_size:%d\n", qscale, (int)diff, br_compensation,
-// rate_q, short_term_q, s->mc_mb_var, s->frame_bits);
-//printf("%d %d\n", s->bit_rate, (int)fps);
- return qscale;
-#else
- INT64 diff, total_bits = s->total_bits;
- float q;
- int qscale;
- if (s->pict_type == I_TYPE) {
- s->wanted_bits += s->I_frame_bits;
- } else {
- s->wanted_bits += s->P_frame_bits;
- }
- diff = s->wanted_bits - total_bits;
- q = 31.0 - (float)diff / (QSCALE_K * s->mb_height * s->mb_width);
- /* adjust for I frame */
- if (s->pict_type == I_TYPE && !s->intra_only) {
- q /= I_FRAME_SIZE_RATIO;
- }
-
- /* using a too small Q scale leeds to problems in mpeg1 and h263
- because AC coefficients are clamped to 255 or 127 */
- qmin = 3;
- if (q < qmin)
- q = qmin;
- else if (q > 31)
- q = 31;
- qscale = (int)(q + 0.5);
-#if defined(DEBUG)
- printf("\n%d: total=%0.0f wanted=%0.0f br=%0.1f diff=%d qest=%2.1f\n",
- s->picture_number,
- (double)total_bits,
- (double)s->wanted_bits,
- (float)s->frame_rate / FRAME_RATE_BASE *
- total_bits / s->picture_number,
- (int)diff, q);
-#endif
- return qscale;
-#endif
+ if(mb_x== s->resync_mb_x && mb_y== s->resync_mb_y) return;
+ if(!s->mbskip_table[mb_x + mb_y*s->mb_width]) mb_dist++;
+ }
+ mb_x=s->mb_width-1;
+ }
}
AVCodec mpeg1video_encoder = {
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-02 22:54:36 +0000