* H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
*
- * This file is part of FFmpeg.
+ * This file is part of Libav.
*
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
- * @file libavcodec/h264.h
+ * @file
* H.264 / AVC / MPEG4 part10 codec.
* @author Michael Niedermayer <michaelni@gmx.at>
*/
#ifndef AVCODEC_H264_H
#define AVCODEC_H264_H
+#include "libavutil/intreadwrite.h"
#include "dsputil.h"
#include "cabac.h"
#include "mpegvideo.h"
+#include "h264dsp.h"
#include "h264pred.h"
+#include "rectangle.h"
#define interlaced_dct interlaced_dct_is_a_bad_name
#define mb_intra mb_intra_is_not_initialized_see_mb_type
-#define LUMA_DC_BLOCK_INDEX 25
-#define CHROMA_DC_BLOCK_INDEX 26
-
-#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
-#define COEFF_TOKEN_VLC_BITS 8
-#define TOTAL_ZEROS_VLC_BITS 9
-#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
-#define RUN_VLC_BITS 3
-#define RUN7_VLC_BITS 6
-
#define MAX_SPS_COUNT 32
#define MAX_PPS_COUNT 256
* of progressive decoding by about 2%. */
#define ALLOW_INTERLACE
-#define ALLOW_NOCHROMA
+#define FMO 0
/**
* The maximum number of slices supported by the decoder.
#define MB_FIELD h->mb_field_decoding_flag
#define FRAME_MBAFF h->mb_aff_frame
#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
+#define LEFT_MBS 2
+#define LTOP 0
+#define LBOT 1
+#define LEFT(i) (i)
#else
#define MB_MBAFF 0
#define MB_FIELD 0
#define FIELD_PICTURE 0
#undef IS_INTERLACED
#define IS_INTERLACED(mb_type) 0
+#define LEFT_MBS 1
+#define LTOP 0
+#define LBOT 0
+#define LEFT(i) 0
#endif
#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
-#ifdef ALLOW_NOCHROMA
-#define CHROMA h->sps.chroma_format_idc
-#else
-#define CHROMA 1
+#ifndef CABAC
+#define CABAC h->pps.cabac
#endif
+#define CHROMA422 (h->sps.chroma_format_idc == 2)
+#define CHROMA444 (h->sps.chroma_format_idc == 3)
+
#define EXTENDED_SAR 255
#define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
#define IS_REF0(a) ((a) & MB_TYPE_REF0)
#define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
+/**
+ * Value of Picture.reference when Picture is not a reference picture, but
+ * is held for delayed output.
+ */
+#define DELAYED_PIC_REF 4
+
+#define QP_MAX_NUM (51 + 2*6) // The maximum supported qp
+
/* NAL unit types */
enum {
NAL_SLICE=1,
int num_reorder_frames;
int scaling_matrix_present;
uint8_t scaling_matrix4[6][16];
- uint8_t scaling_matrix8[2][64];
+ uint8_t scaling_matrix8[6][64];
int nal_hrd_parameters_present_flag;
int vcl_hrd_parameters_present_flag;
int pic_struct_present_flag;
int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
int residual_color_transform_flag; ///< residual_colour_transform_flag
+ int constraint_set_flags; ///< constraint_set[0-3]_flag
}SPS;
/**
int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
int transform_8x8_mode; ///< transform_8x8_mode_flag
uint8_t scaling_matrix4[6][16];
- uint8_t scaling_matrix8[2][64];
+ uint8_t scaling_matrix8[6][64];
uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
int chroma_qp_diff;
}PPS;
*/
typedef struct H264Context{
MpegEncContext s;
- int nal_ref_idc;
- int nal_unit_type;
- uint8_t *rbsp_buffer[2];
- unsigned int rbsp_buffer_size[2];
-
- /**
- * Used to parse AVC variant of h264
- */
- int is_avc; ///< this flag is != 0 if codec is avc1
- int got_avcC; ///< flag used to parse avcC data only once
- int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
-
+ H264DSPContext h264dsp;
+ int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
int chroma_qp[2]; //QPc
+ int qp_thresh; ///< QP threshold to skip loopfilter
+
int prev_mb_skipped;
int next_mb_skipped;
int chroma_pred_mode;
int intra16x16_pred_mode;
+ int topleft_mb_xy;
int top_mb_xy;
- int left_mb_xy[2];
+ int topright_mb_xy;
+ int left_mb_xy[LEFT_MBS];
+
+ int topleft_type;
+ int top_type;
+ int topright_type;
+ int left_type[LEFT_MBS];
+
+ const uint8_t * left_block;
+ int topleft_partition;
int8_t intra4x4_pred_mode_cache[5*8];
- int8_t (*intra4x4_pred_mode)[8];
+ int8_t (*intra4x4_pred_mode);
H264PredContext hpc;
unsigned int topleft_samples_available;
unsigned int top_samples_available;
unsigned int topright_samples_available;
unsigned int left_samples_available;
- uint8_t (*top_borders[2])[16+2*8];
- uint8_t left_border[2*(17+2*9)];
+ uint8_t (*top_borders[2])[(16*3)*2];
/**
* non zero coeff count cache.
* is 64 if not available.
*/
- DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
- uint8_t (*non_zero_count)[16];
+ DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15*8];
+
+ uint8_t (*non_zero_count)[48];
/**
* Motion vector cache.
*/
- DECLARE_ALIGNED_8(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
- /**
- * is 1 if the specific list MV&references are set to 0,0,-2.
- */
- int mv_cache_clean[2];
-
/**
* number of neighbors (top and/or left) that used 8x8 dct
*/
* block_offset[ 0..23] for frame macroblocks
* block_offset[24..47] for field macroblocks
*/
- int block_offset[2*(16+8)];
+ int block_offset[2*(16*3)];
uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
- uint32_t *mb2b8_xy;
+ uint32_t *mb2br_xy;
int b_stride; //FIXME use s->b4_stride
- int b8_stride;
int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
int mb_uvlinesize;
int emu_edge_width;
int emu_edge_height;
- int halfpel_flag;
- int thirdpel_flag;
-
- int unknown_svq3_flag;
- int next_slice_index;
-
- SPS *sps_buffers[MAX_SPS_COUNT];
SPS sps; ///< current sps
- PPS *pps_buffers[MAX_PPS_COUNT];
/**
* current pps
*/
PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
- uint32_t dequant4_buffer[6][52][16];
- uint32_t dequant8_buffer[2][52][64];
+ uint32_t dequant4_buffer[6][QP_MAX_NUM+1][16]; //FIXME should these be moved down?
+ uint32_t dequant8_buffer[6][QP_MAX_NUM+1][64];
uint32_t (*dequant4_coeff[6])[16];
- uint32_t (*dequant8_coeff[2])[64];
- int dequant_coeff_pps; ///< reinit tables when pps changes
+ uint32_t (*dequant8_coeff[6])[64];
int slice_num;
- uint16_t *slice_table_base;
uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
int slice_type;
int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
int mb_field_decoding_flag;
int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
- DECLARE_ALIGNED_8(uint16_t, sub_mb_type[4]);
-
- //POC stuff
- int poc_lsb;
- int poc_msb;
- int delta_poc_bottom;
- int delta_poc[2];
- int frame_num;
- int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
- int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
- int frame_num_offset; ///< for POC type 2
- int prev_frame_num_offset; ///< for POC type 2
- int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
-
- /**
- * frame_num for frames or 2*frame_num+1 for field pics.
- */
- int curr_pic_num;
-
- /**
- * max_frame_num or 2*max_frame_num for field pics.
- */
- int max_pic_num;
+ DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
//Weighted pred stuff
int use_weight;
int use_weight_chroma;
int luma_log2_weight_denom;
int chroma_log2_weight_denom;
- int luma_weight[2][48];
- int luma_offset[2][48];
- int chroma_weight[2][48][2];
- int chroma_offset[2][48][2];
- int implicit_weight[48][48];
-
- //deblock
- int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
- int slice_alpha_c0_offset;
- int slice_beta_offset;
-
- int redundant_pic_count;
+ //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
+ 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;
+ int col_fieldoff;
int dist_scale_factor[16];
int dist_scale_factor_field[2][32];
int map_col_to_list0[2][16+32];
*/
unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
unsigned int list_count;
- Picture *short_ref[32];
- Picture *long_ref[32];
- Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
+ uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
Reordered version of default_ref_list
according to picture reordering in slice header */
int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
- Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
- int outputed_poc;
-
- /**
- * memory management control operations buffer.
- */
- MMCO mmco[MAX_MMCO_COUNT];
- int mmco_index;
-
- int long_ref_count; ///< number of actual long term references
- int short_ref_count; ///< number of actual short term references
//data partitioning
GetBitContext intra_gb;
GetBitContext *intra_gb_ptr;
GetBitContext *inter_gb_ptr;
- 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
+ DECLARE_ALIGNED(16, DCTELEM, mb)[16*48*2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
+ DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16*2];
+ DCTELEM mb_padding[256*2]; ///< 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
/**
* Cabac
*/
CABACContext cabac;
- uint8_t cabac_state[460];
- int cabac_init_idc;
+ uint8_t cabac_state[1024];
/* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
uint16_t *cbp_table;
/* chroma_pred_mode for i4x4 or i16x16, else 0 */
uint8_t *chroma_pred_mode_table;
int last_qscale_diff;
- int16_t (*mvd_table[2])[2];
- DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
+ uint8_t (*mvd_table[2])[2];
+ DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
uint8_t *direct_table;
uint8_t direct_cache[5*8];
int x264_build;
+ int mb_xy;
+
+ int is_complex;
+
+ //deblock
+ int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
+ int slice_alpha_c0_offset;
+ int slice_beta_offset;
+
+//=============================================================
+ //Things below are not used in the MB or more inner code
+
+ int nal_ref_idc;
+ int nal_unit_type;
+ uint8_t *rbsp_buffer[2];
+ unsigned int rbsp_buffer_size[2];
+
+ /**
+ * Used to parse AVC variant of h264
+ */
+ 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 dequant_coeff_pps; ///< reinit tables when pps changes
+
+ uint16_t *slice_table_base;
+
+
+ //POC stuff
+ int poc_lsb;
+ int poc_msb;
+ int delta_poc_bottom;
+ int delta_poc[2];
+ int frame_num;
+ int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
+ int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
+ int frame_num_offset; ///< for POC type 2
+ int prev_frame_num_offset; ///< for POC type 2
+ int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
+
+ /**
+ * frame_num for frames or 2*frame_num+1 for field pics.
+ */
+ int curr_pic_num;
+
+ /**
+ * max_frame_num or 2*max_frame_num for field pics.
+ */
+ int max_pic_num;
+
+ int redundant_pic_count;
+
+ Picture *short_ref[32];
+ Picture *long_ref[32];
+ Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
+ Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
+ int last_pocs[MAX_DELAYED_PIC_COUNT];
+ Picture *next_output_pic;
+ int outputed_poc;
+ int next_outputed_poc;
+
/**
- * @defgroup multithreading Members for slice based multithreading
+ * memory management control operations buffer.
+ */
+ MMCO mmco[MAX_MMCO_COUNT];
+ int mmco_index;
+ int mmco_reset;
+
+ int long_ref_count; ///< number of actual long term references
+ int short_ref_count; ///< number of actual short term references
+
+ int cabac_init_idc;
+
+ /**
+ * @name Members for slice based multithreading
* @{
*/
struct H264Context *thread_context[MAX_THREADS];
int last_slice_type;
/** @} */
- int mb_xy;
-
- uint32_t svq3_watermark_key;
-
/**
* pic_struct in picture timing SEI message
*/
*/
int sei_recovery_frame_cnt;
- int is_complex;
-
int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
// Timestamp stuff
int sei_buffering_period_present; ///< Buffering period SEI flag
int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
-}H264Context;
+ int cur_chroma_format_idc;
+}H264Context;
-extern const uint8_t ff_h264_chroma_qp[52];
+extern const uint8_t ff_h264_chroma_qp[3][QP_MAX_NUM+1]; ///< One chroma qp table for each supported bit depth (8, 9, 10).
+extern const uint16_t ff_h264_mb_sizes[4];
/**
* Decode SEI
*/
int ff_h264_decode_seq_parameter_set(H264Context *h);
+/**
+ * compute profile from sps
+ */
+int ff_h264_get_profile(SPS *sps);
+
/**
* Decode PPS
*/
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
- * @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(H264Context *h);
+int ff_h264_get_slice_type(const H264Context *h);
/**
- * allocates tables.
+ * Allocate tables.
* needs width/height
*/
int ff_h264_alloc_tables(H264Context *h);
/**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Fill the default_ref_list.
*/
-int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
+int ff_h264_fill_default_ref_list(H264Context *h);
+
+int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
+void ff_h264_fill_mbaff_ref_list(H264Context *h);
+void ff_h264_remove_all_refs(H264Context *h);
+
+/**
+ * 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);
+
+
+/**
+ * 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);
+
+/**
+ * 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, int is_chroma);
-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);
+
+/**
+ * Decode a macroblock
+ * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR if an error is noticed
+ */
+int ff_h264_decode_mb_cavlc(H264Context *h);
+
+/**
+ * Decode a CABAC coded macroblock
+ * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR if an error is noticed
+ */
+int ff_h264_decode_mb_cabac(H264Context *h);
+
+void ff_h264_init_cabac_states(H264Context *h);
void ff_h264_direct_dist_scale_factor(H264Context * const h);
void ff_h264_direct_ref_list_init(H264Context * const h);
/ / /
o-o o-o
*/
+
+/* Scan8 organization:
+ * 0 1 2 3 4 5 6 7
+ * 0 DY y y y y y
+ * 1 y Y Y Y Y
+ * 2 y Y Y Y Y
+ * 3 y Y Y Y Y
+ * 4 y Y Y Y Y
+ * 5 DU u u u u u
+ * 6 u U U U U
+ * 7 u U U U U
+ * 8 u U U U U
+ * 9 u U U U U
+ * 10 DV v v v v v
+ * 11 v V V V V
+ * 12 v V V V V
+ * 13 v V V V V
+ * 14 v V V V V
+ * DY/DU/DV are for luma/chroma DC.
+ */
+
+#define LUMA_DC_BLOCK_INDEX 48
+#define CHROMA_DC_BLOCK_INDEX 49
+
//This table must be here because scan8[constant] must be known at compiletime
-static const uint8_t scan8[16 + 2*4]={
- 4+1*8, 5+1*8, 4+2*8, 5+2*8,
- 6+1*8, 7+1*8, 6+2*8, 7+2*8,
- 4+3*8, 5+3*8, 4+4*8, 5+4*8,
- 6+3*8, 7+3*8, 6+4*8, 7+4*8,
- 1+1*8, 2+1*8,
- 1+2*8, 2+2*8,
- 1+4*8, 2+4*8,
- 1+5*8, 2+5*8,
+static const uint8_t scan8[16*3 + 3]={
+ 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
+ 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
+ 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
+ 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
+ 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
+ 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
+ 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
+ 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
+ 4+11*8, 5+11*8, 4+12*8, 5+12*8,
+ 6+11*8, 7+11*8, 6+12*8, 7+12*8,
+ 4+13*8, 5+13*8, 4+14*8, 5+14*8,
+ 6+13*8, 7+13*8, 6+14*8, 7+14*8,
+ 0+ 0*8, 0+ 5*8, 0+10*8
};
static av_always_inline uint32_t pack16to32(int a, int b){
#endif
}
+static av_always_inline uint16_t pack8to16(int a, int b){
+#if HAVE_BIGENDIAN
+ return (b&0xFF) + (a<<8);
+#else
+ return (a&0xFF) + (b<<8);
+#endif
+}
+
/**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Get the chroma qp.
*/
-static inline int check_intra4x4_pred_mode(H264Context *h){
- MpegEncContext * const s = &h->s;
- static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
- static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
- int i;
-
- if(!(h->top_samples_available&0x8000)){
- for(i=0; i<4; i++){
- int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
- if(status<0){
- av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
- return -1;
- } else if(status){
- h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
- }
- }
- }
-
- if((h->left_samples_available&0x8888)!=0x8888){
- static const int mask[4]={0x8000,0x2000,0x80,0x20};
- for(i=0; i<4; i++){
- if(!(h->left_samples_available&mask[i])){
- int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
- if(status<0){
- av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
- return -1;
- } else if(status){
- h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
- }
- }
- }
- }
-
- return 0;
-} //FIXME cleanup like ff_h264_check_intra_pred_mode
+static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
+ return h->pps.chroma_qp_table[t][qscale];
+}
/**
- * gets the chroma qp.
+ * Get the predicted intra4x4 prediction mode.
*/
-static inline int get_chroma_qp(H264Context *h, int t, int qscale){
- return h->pps.chroma_qp_table[t][qscale];
+static av_always_inline int pred_intra_mode(H264Context *h, int n){
+ const int index8= scan8[n];
+ const int left= h->intra4x4_pred_mode_cache[index8 - 1];
+ const int top = h->intra4x4_pred_mode_cache[index8 - 8];
+ const int min= FFMIN(left, top);
+
+ tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
+
+ if(min<0) return DC_PRED;
+ else return min;
}
-static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
- const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
- MpegEncContext *s = &h->s;
-
- /* there is no consistent mapping of mvs to neighboring locations that will
- * make mbaff happy, so we can't move all this logic to fill_caches */
- if(FRAME_MBAFF){
- const uint32_t *mb_types = s->current_picture_ptr->mb_type;
- const int16_t *mv;
- *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
- *C = h->mv_cache[list][scan8[0]-2];
-
- if(!MB_FIELD
- && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
- int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
- if(IS_INTERLACED(mb_types[topright_xy])){
-#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
- const int x4 = X4, y4 = Y4;\
- const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
- if(!USES_LIST(mb_type,list))\
- return LIST_NOT_USED;\
- mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
- h->mv_cache[list][scan8[0]-2][0] = mv[0];\
- h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
- return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
-
- SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
- }
- }
- if(topright_ref == PART_NOT_AVAILABLE
- && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
- && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
- if(!MB_FIELD
- && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
- SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
- }
- if(MB_FIELD
- && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
- && i >= scan8[0]+8){
- // left shift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's OK.
- SET_DIAG_MV(/2, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
- }
- }
-#undef SET_DIAG_MV
+static av_always_inline void write_back_intra_pred_mode(H264Context *h){
+ int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
+ int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
+
+ AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
+ i4x4[4]= i4x4_cache[7+8*3];
+ i4x4[5]= i4x4_cache[7+8*2];
+ i4x4[6]= i4x4_cache[7+8*1];
+}
+
+static av_always_inline void write_back_non_zero_count(H264Context *h){
+ const int mb_xy= h->mb_xy;
+ uint8_t *nnz = h->non_zero_count[mb_xy];
+ uint8_t *nnz_cache = h->non_zero_count_cache;
+
+ AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
+ AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
+ AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
+ AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
+ AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
+ AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
+ AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
+ AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
+
+ if(!h->s.chroma_y_shift){
+ AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
+ AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
+ AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
+ AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
}
+}
- if(topright_ref != PART_NOT_AVAILABLE){
- *C= h->mv_cache[list][ i - 8 + part_width ];
- return topright_ref;
- }else{
- tprintf(s->avctx, "topright MV not available\n");
+static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
+ int b_xy, int b8_xy, int mb_type, int list )
+{
+ int16_t (*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
+ int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
+ AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
+ AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
+ AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
+ AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
+ if( CABAC ) {
+ uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
+ uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
+ if(IS_SKIP(mb_type))
+ AV_ZERO128(mvd_dst);
+ else{
+ AV_COPY64(mvd_dst, mvd_src + 8*3);
+ AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
+ AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
+ AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
+ }
+ }
- *C= h->mv_cache[list][ i - 8 - 1 ];
- return h->ref_cache[list][ i - 8 - 1 ];
+ {
+ int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
+ int8_t *ref_cache = h->ref_cache[list];
+ ref_index[0+0*2]= ref_cache[scan8[0]];
+ ref_index[1+0*2]= ref_cache[scan8[4]];
+ ref_index[0+1*2]= ref_cache[scan8[8]];
+ ref_index[1+1*2]= ref_cache[scan8[12]];
}
}
-/**
- * gets the predicted MV.
- * @param n the block index
- * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
- * @param mx the x component of the predicted motion vector
- * @param my the y component of the predicted motion vector
- */
-static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
- const int index8= scan8[n];
- const int top_ref= h->ref_cache[list][ index8 - 8 ];
- const int left_ref= h->ref_cache[list][ index8 - 1 ];
- const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
- const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
- const int16_t * C;
- int diagonal_ref, match_count;
-
- assert(part_width==1 || part_width==2 || part_width==4);
-
-/* mv_cache
- B . . A T T T T
- U . . L . . , .
- U . . L . . . .
- U . . L . . , .
- . . . L . . . .
-*/
+static av_always_inline void write_back_motion(H264Context *h, int mb_type){
+ MpegEncContext * const s = &h->s;
+ const int b_stride = h->b_stride;
+ const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
+ const int b8_xy= 4*h->mb_xy;
- diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
- match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
- tprintf(h->s.avctx, "pred_motion match_count=%d\n", match_count);
- if(match_count > 1){ //most common
- *mx= mid_pred(A[0], B[0], C[0]);
- *my= mid_pred(A[1], B[1], C[1]);
- }else if(match_count==1){
- if(left_ref==ref){
- *mx= A[0];
- *my= A[1];
- }else if(top_ref==ref){
- *mx= B[0];
- *my= B[1];
- }else{
- *mx= C[0];
- *my= C[1];
- }
+ if(USES_LIST(mb_type, 0)){
+ write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
}else{
- if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
- *mx= A[0];
- *my= A[1];
- }else{
- *mx= mid_pred(A[0], B[0], C[0]);
- *my= mid_pred(A[1], B[1], C[1]);
- }
+ fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
+ 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
+ }
+ if(USES_LIST(mb_type, 1)){
+ write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
}
- tprintf(h->s.avctx, "pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
+ if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
+ if(IS_8X8(mb_type)){
+ uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
+ direct_table[1] = h->sub_mb_type[1]>>1;
+ direct_table[2] = h->sub_mb_type[2]>>1;
+ direct_table[3] = h->sub_mb_type[3]>>1;
+ }
+ }
}
+static av_always_inline int get_dct8x8_allowed(H264Context *h){
+ if(h->sps.direct_8x8_inference_flag)
+ return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
+ else
+ return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
+}
#endif /* AVCODEC_H264_H */