*/
/**
- * @file libavcodec/h264.h
+ * @file
* H.264 / AVC / MPEG4 part10 codec.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
#include "dsputil.h"
#include "cabac.h"
#include "mpegvideo.h"
+#include "h264dsp.h"
#include "h264pred.h"
#include "rectangle.h"
*/
typedef struct H264Context{
MpegEncContext s;
+ H264DSPContext h264dsp;
int chroma_qp[2]; //QPc
int qp_thresh; ///< QP threshold to skip loopfilter
* non zero coeff count cache.
* is 64 if not available.
*/
- DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
+ DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
/*
.UU.YYYY
/**
* Motion vector cache.
*/
- DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
- DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
+ DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
+ DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
#define LIST_NOT_USED -1 //FIXME rename?
#define PART_NOT_AVAILABLE -2
int mb_field_decoding_flag;
int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
- DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
+ DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
//Weighted pred stuff
int use_weight;
int luma_log2_weight_denom;
int chroma_log2_weight_denom;
//The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
- int luma_weight[2][48][2];
- int chroma_weight[2][48][2][2];
- int implicit_weight[48][48];
+ int luma_weight[48][2][2];
+ int chroma_weight[48][2][2][2];
+ int implicit_weight[48][48][2];
int direct_spatial_mv_pred;
int col_parity;
GetBitContext *intra_gb_ptr;
GetBitContext *inter_gb_ptr;
- DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
+ DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
DCTELEM mb_padding[256]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
/**
uint8_t *chroma_pred_mode_table;
int last_qscale_diff;
uint8_t (*mvd_table[2])[2];
- DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
+ DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
uint8_t *direct_table;
uint8_t direct_cache[5*8];
*/
int is_avc; ///< this flag is != 0 if codec is avc1
int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
+ int got_first; ///< this flag is != 0 if we've parsed a frame
SPS *sps_buffers[MAX_SPS_COUNT];
PPS *pps_buffers[MAX_PPS_COUNT];
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
/**
- * Decodes a network abstraction layer unit.
+ * Decode a network abstraction layer unit.
* @param consumed is the number of bytes used as input
* @param length is the length of the array
* @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
- * @returns decoded bytes, might be src+1 if no escapes
+ * @return decoded bytes, might be src+1 if no escapes
*/
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
/**
- * identifies the exact end of the bitstream
+ * Identify the exact end of the bitstream
* @return the length of the trailing, or 0 if damaged
*/
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
/**
- * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
+ * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
*/
av_cold void ff_h264_free_context(H264Context *h);
/**
- * reconstructs bitstream slice_type.
+ * Reconstruct bitstream slice_type.
*/
int ff_h264_get_slice_type(const H264Context *h);
/**
- * allocates tables.
+ * Allocate tables.
* needs width/height
*/
int ff_h264_alloc_tables(H264Context *h);
/**
- * fills the default_ref_list.
+ * Fill the default_ref_list.
*/
int ff_h264_fill_default_ref_list(H264Context *h);
void ff_h264_remove_all_refs(H264Context *h);
/**
- * Executes the reference picture marking (memory management control operations).
+ * Execute the reference picture marking (memory management control operations).
*/
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
+void ff_generate_sliding_window_mmcos(H264Context *h);
+
/**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
*/
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
/**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
*/
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
void ff_h264_write_back_intra_pred_mode(H264Context *h);
void ff_h264_hl_decode_mb(H264Context *h);
int ff_h264_frame_start(H264Context *h);
+int ff_h264_decode_extradata(H264Context *h);
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
av_cold void ff_h264_decode_init_vlc(void);
/**
- * decodes a macroblock
- * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
+ * Decode a macroblock
+ * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
*/
int ff_h264_decode_mb_cavlc(H264Context *h);
/**
- * decodes a CABAC coded macroblock
- * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
+ * Decode a CABAC coded macroblock
+ * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
*/
int ff_h264_decode_mb_cabac(H264Context *h);
h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
+ if(FMO){
if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
- if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
+ }else{
+ if(h->slice_table[topleft_xy ] != h->slice_num){
+ h->topleft_type = 0;
+ if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
+ if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
+ }
+ }
+ if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
}
static void fill_decode_caches(H264Context *h, int mb_type){
fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
if(IS_DIRECT(top_type)){
- AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
+ AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
}else if(IS_8X8(top_type)){
int b8_xy = 4*top_xy;
h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
}
-/**
- *
- * @returns non zero if the loop filter can be skiped
- */
-static int fill_filter_caches(H264Context *h, int mb_type){
- MpegEncContext * const s = &h->s;
- const int mb_xy= h->mb_xy;
- int top_xy, left_xy[2];
- int top_type, left_type[2];
-
- top_xy = mb_xy - (s->mb_stride << MB_FIELD);
-
- //FIXME deblocking could skip the intra and nnz parts.
-
- /* Wow, what a mess, why didn't they simplify the interlacing & intra
- * stuff, I can't imagine that these complex rules are worth it. */
-
- left_xy[1] = left_xy[0] = mb_xy-1;
- if(FRAME_MBAFF){
- const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
- const int curr_mb_field_flag = IS_INTERLACED(mb_type);
- if(s->mb_y&1){
- if (left_mb_field_flag != curr_mb_field_flag) {
- left_xy[0] -= s->mb_stride;
- }
- }else{
- if(curr_mb_field_flag){
- top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
- }
- if (left_mb_field_flag != curr_mb_field_flag) {
- left_xy[1] += s->mb_stride;
- }
- }
- }
-
- h->top_mb_xy = top_xy;
- h->left_mb_xy[0] = left_xy[0];
- h->left_mb_xy[1] = left_xy[1];
- {
- //for sufficiently low qp, filtering wouldn't do anything
- //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
- int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
- int qp = s->current_picture.qscale_table[mb_xy];
- if(qp <= qp_thresh
- && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
- && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
- if(!FRAME_MBAFF)
- return 1;
- if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
- && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
- return 1;
- }
- }
-
- if(h->deblocking_filter == 2){
- h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
- h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
- h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
- }else{
- h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
- h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
- h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
- }
- if(IS_INTRA(mb_type))
- return 0;
-
- AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
- AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
- AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
- AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
- AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
-
- h->cbp= h->cbp_table[mb_xy];
-
- {
- int list;
- for(list=0; list<h->list_count; list++){
- int8_t *ref;
- int y, b_stride;
- int16_t (*mv_dst)[2];
- int16_t (*mv_src)[2];
-
- if(!USES_LIST(mb_type, list)){
- fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
- AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
- AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
- AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
- AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
- continue;
- }
-
- ref = &s->current_picture.ref_index[list][4*mb_xy];
- {
- int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
- AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
- AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
- ref += 2;
- AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
- AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
- }
-
- b_stride = h->b_stride;
- mv_dst = &h->mv_cache[list][scan8[0]];
- mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
- for(y=0; y<4; y++){
- AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
- }
-
- }
- }
-
-
-/*
-0 . T T. T T T T
-1 L . .L . . . .
-2 L . .L . . . .
-3 . T TL . . . .
-4 L . .L . . . .
-5 L . .. . . . .
-*/
-//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
- if(top_type){
- AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
- }
-
- if(left_type[0]){
- h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
- h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
- h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
- h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
- }
-
- // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
- if(!CABAC && h->pps.transform_8x8_mode){
- if(IS_8x8DCT(top_type)){
- h->non_zero_count_cache[4+8*0]=
- h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
- h->non_zero_count_cache[6+8*0]=
- h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
- }
- if(IS_8x8DCT(left_type[0])){
- h->non_zero_count_cache[3+8*1]=
- h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
- }
- if(IS_8x8DCT(left_type[1])){
- h->non_zero_count_cache[3+8*3]=
- h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
- }
-
- if(IS_8x8DCT(mb_type)){
- h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
- h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
-
- h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
- h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
-
- h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
- h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
-
- h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
- h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
- }
- }
-
- if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
- int list;
- for(list=0; list<h->list_count; list++){
- if(USES_LIST(top_type, list)){
- const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
- const int b8_xy= 4*top_xy + 2;
- int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
- AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
- h->ref_cache[list][scan8[0] + 0 - 1*8]=
- h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
- h->ref_cache[list][scan8[0] + 2 - 1*8]=
- h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
- }else{
- AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
- AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
- }
-
- if(!IS_INTERLACED(mb_type^left_type[0])){
- if(USES_LIST(left_type[0], list)){
- const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
- const int b8_xy= 4*left_xy[0] + 1;
- int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
- AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
- AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
- AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
- AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
- h->ref_cache[list][scan8[0] - 1 + 0 ]=
- h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
- h->ref_cache[list][scan8[0] - 1 +16 ]=
- h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
- }else{
- AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
- AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
- AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
- AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
- h->ref_cache[list][scan8[0] - 1 + 0 ]=
- h->ref_cache[list][scan8[0] - 1 + 8 ]=
- h->ref_cache[list][scan8[0] - 1 + 16 ]=
- h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
- }
- }
- }
- }
-
- return 0;
-}
-
/**
* gets the predicted intra4x4 prediction mode.
*/
/**
* decodes a P_SKIP or B_SKIP macroblock
*/
-static void decode_mb_skip(H264Context *h){
+static void av_unused decode_mb_skip(H264Context *h){
MpegEncContext * const s = &h->s;
const int mb_xy= h->mb_xy;
int mb_type=0;
projects / ffmpeg / blobdiff