Use pointers to avoid copying AVFrame.
[libav.git] / libavcodec / h264.c
CommitLineData
0da71265
MN
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
b78e7197
DB
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
0da71265
MN
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
b78e7197 10 * version 2.1 of the License, or (at your option) any later version.
0da71265 11 *
b78e7197 12 * FFmpeg is distributed in the hope that it will be useful,
0da71265
MN
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
b78e7197 18 * License along with FFmpeg; if not, write to the Free Software
5509bffa 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0da71265
MN
20 *
21 */
115329f1 22
0da71265
MN
23/**
24 * @file h264.c
25 * H.264 / AVC / MPEG4 part10 codec.
26 * @author Michael Niedermayer <michaelni@gmx.at>
27 */
28
29#include "common.h"
30#include "dsputil.h"
31#include "avcodec.h"
32#include "mpegvideo.h"
33#include "h264data.h"
34#include "golomb.h"
35
e5017ab8
LA
36#include "cabac.h"
37
2848ce84 38//#undef NDEBUG
0da71265
MN
39#include <assert.h>
40
41#define interlaced_dct interlaced_dct_is_a_bad_name
42#define mb_intra mb_intra_isnt_initalized_see_mb_type
43
44#define LUMA_DC_BLOCK_INDEX 25
45#define CHROMA_DC_BLOCK_INDEX 26
46
47#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
48#define COEFF_TOKEN_VLC_BITS 8
49#define TOTAL_ZEROS_VLC_BITS 9
50#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
51#define RUN_VLC_BITS 3
52#define RUN7_VLC_BITS 6
53
54#define MAX_SPS_COUNT 32
55#define MAX_PPS_COUNT 256
56
57#define MAX_MMCO_COUNT 66
58
5d18eaad
LM
59/* Compiling in interlaced support reduces the speed
60 * of progressive decoding by about 2%. */
61#define ALLOW_INTERLACE
62
63#ifdef ALLOW_INTERLACE
64#define MB_MBAFF h->mb_mbaff
65#define MB_FIELD h->mb_field_decoding_flag
66#define FRAME_MBAFF h->mb_aff_frame
67#else
68#define MB_MBAFF 0
69#define MB_FIELD 0
70#define FRAME_MBAFF 0
71#undef IS_INTERLACED
72#define IS_INTERLACED(mb_type) 0
73#endif
74
0da71265
MN
75/**
76 * Sequence parameter set
77 */
78typedef struct SPS{
115329f1 79
0da71265
MN
80 int profile_idc;
81 int level_idc;
36940eca 82 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
0da71265
MN
83 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
84 int poc_type; ///< pic_order_cnt_type
85 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
86 int delta_pic_order_always_zero_flag;
87 int offset_for_non_ref_pic;
88 int offset_for_top_to_bottom_field;
89 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
90 int ref_frame_count; ///< num_ref_frames
a15e68de 91 int gaps_in_frame_num_allowed_flag;
0da71265
MN
92 int mb_width; ///< frame_width_in_mbs_minus1 + 1
93 int mb_height; ///< frame_height_in_mbs_minus1 + 1
94 int frame_mbs_only_flag;
95 int mb_aff; ///<mb_adaptive_frame_field_flag
96 int direct_8x8_inference_flag;
a15e68de
MN
97 int crop; ///< frame_cropping_flag
98 int crop_left; ///< frame_cropping_rect_left_offset
99 int crop_right; ///< frame_cropping_rect_right_offset
100 int crop_top; ///< frame_cropping_rect_top_offset
101 int crop_bottom; ///< frame_cropping_rect_bottom_offset
0da71265 102 int vui_parameters_present_flag;
5ff85f1d 103 AVRational sar;
ec587a50
MR
104 int timing_info_present_flag;
105 uint32_t num_units_in_tick;
106 uint32_t time_scale;
107 int fixed_frame_rate_flag;
0da71265 108 short offset_for_ref_frame[256]; //FIXME dyn aloc?
4e4d983e
LM
109 int bitstream_restriction_flag;
110 int num_reorder_frames;
239ea04c
LM
111 int scaling_matrix_present;
112 uint8_t scaling_matrix4[6][16];
113 uint8_t scaling_matrix8[2][64];
0da71265
MN
114}SPS;
115
116/**
117 * Picture parameter set
118 */
119typedef struct PPS{
ac658be5 120 unsigned int sps_id;
0da71265
MN
121 int cabac; ///< entropy_coding_mode_flag
122 int pic_order_present; ///< pic_order_present_flag
123 int slice_group_count; ///< num_slice_groups_minus1 + 1
124 int mb_slice_group_map_type;
88e7a4d1 125 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
0da71265
MN
126 int weighted_pred; ///< weighted_pred_flag
127 int weighted_bipred_idc;
128 int init_qp; ///< pic_init_qp_minus26 + 26
129 int init_qs; ///< pic_init_qs_minus26 + 26
130 int chroma_qp_index_offset;
131 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
132 int constrained_intra_pred; ///< constrained_intra_pred_flag
133 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
43efd19a 134 int transform_8x8_mode; ///< transform_8x8_mode_flag
239ea04c
LM
135 uint8_t scaling_matrix4[6][16];
136 uint8_t scaling_matrix8[2][64];
0da71265
MN
137}PPS;
138
139/**
140 * Memory management control operation opcode.
141 */
142typedef enum MMCOOpcode{
143 MMCO_END=0,
144 MMCO_SHORT2UNUSED,
145 MMCO_LONG2UNUSED,
146 MMCO_SHORT2LONG,
147 MMCO_SET_MAX_LONG,
115329f1 148 MMCO_RESET,
0da71265
MN
149 MMCO_LONG,
150} MMCOOpcode;
151
152/**
153 * Memory management control operation.
154 */
155typedef struct MMCO{
156 MMCOOpcode opcode;
157 int short_frame_num;
158 int long_index;
159} MMCO;
160
161/**
162 * H264Context
163 */
164typedef struct H264Context{
165 MpegEncContext s;
115329f1 166 int nal_ref_idc;
0da71265 167 int nal_unit_type;
0da71265 168 uint8_t *rbsp_buffer;
f038fe8b 169 unsigned int rbsp_buffer_size;
0da71265 170
4770b1b4
RT
171 /**
172 * Used to parse AVC variant of h264
173 */
174 int is_avc; ///< this flag is != 0 if codec is avc1
175 int got_avcC; ///< flag used to parse avcC data only once
176 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
177
0da71265
MN
178 int chroma_qp; //QPc
179
5d18eaad
LM
180 int prev_mb_skipped;
181 int next_mb_skipped;
0da71265
MN
182
183 //prediction stuff
184 int chroma_pred_mode;
185 int intra16x16_pred_mode;
6ba71fc4
LLL
186
187 int top_mb_xy;
188 int left_mb_xy[2];
115329f1 189
0da71265
MN
190 int8_t intra4x4_pred_mode_cache[5*8];
191 int8_t (*intra4x4_pred_mode)[8];
192 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
43efd19a 193 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
0da71265
MN
194 void (*pred8x8 [4+3])(uint8_t *src, int stride);
195 void (*pred16x16[4+3])(uint8_t *src, int stride);
196 unsigned int topleft_samples_available;
197 unsigned int top_samples_available;
198 unsigned int topright_samples_available;
199 unsigned int left_samples_available;
6ba71fc4
LLL
200 uint8_t (*top_borders[2])[16+2*8];
201 uint8_t left_border[2*(17+2*9)];
0da71265
MN
202
203 /**
204 * non zero coeff count cache.
205 * is 64 if not available.
206 */
68b51e58 207 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
53c05b1e 208 uint8_t (*non_zero_count)[16];
0da71265
MN
209
210 /**
211 * Motion vector cache.
212 */
68b51e58
SH
213 DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
214 DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
0da71265
MN
215#define LIST_NOT_USED -1 //FIXME rename?
216#define PART_NOT_AVAILABLE -2
115329f1 217
0da71265
MN
218 /**
219 * is 1 if the specific list MV&references are set to 0,0,-2.
220 */
221 int mv_cache_clean[2];
222
6867a90b 223 /**
43efd19a
LM
224 * number of neighbors (top and/or left) that used 8x8 dct
225 */
226 int neighbor_transform_size;
227
228 /**
6867a90b
LLL
229 * block_offset[ 0..23] for frame macroblocks
230 * block_offset[24..47] for field macroblocks
231 */
232 int block_offset[2*(16+8)];
115329f1 233
a55f20bd
LM
234 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
235 uint32_t *mb2b8_xy;
650cec0c 236 int b_stride; //FIXME use s->b4_stride
0da71265
MN
237 int b8_stride;
238
5d18eaad
LM
239 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
240 int mb_uvlinesize;
241
242 int emu_edge_width;
243 int emu_edge_height;
244
8b82a956
MN
245 int halfpel_flag;
246 int thirdpel_flag;
247
da3b9756
MM
248 int unknown_svq3_flag;
249 int next_slice_index;
250
0da71265
MN
251 SPS sps_buffer[MAX_SPS_COUNT];
252 SPS sps; ///< current sps
115329f1 253
0da71265
MN
254 PPS pps_buffer[MAX_PPS_COUNT];
255 /**
256 * current pps
257 */
3b66c4c5 258 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
0da71265 259
239ea04c
LM
260 uint32_t dequant4_buffer[6][52][16];
261 uint32_t dequant8_buffer[2][52][64];
262 uint32_t (*dequant4_coeff[6])[16];
263 uint32_t (*dequant8_coeff[2])[64];
264 int dequant_coeff_pps; ///< reinit tables when pps changes
43efd19a 265
0da71265
MN
266 int slice_num;
267 uint8_t *slice_table_base;
5d18eaad 268 uint8_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
0da71265
MN
269 int slice_type;
270 int slice_type_fixed;
115329f1 271
0da71265 272 //interlacing specific flags
6ba71fc4 273 int mb_aff_frame;
0da71265 274 int mb_field_decoding_flag;
5d18eaad 275 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
115329f1 276
88e7a4d1 277 unsigned int sub_mb_type[4];
115329f1 278
0da71265
MN
279 //POC stuff
280 int poc_lsb;
281 int poc_msb;
282 int delta_poc_bottom;
283 int delta_poc[2];
284 int frame_num;
285 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
286 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
287 int frame_num_offset; ///< for POC type 2
288 int prev_frame_num_offset; ///< for POC type 2
289 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
290
291 /**
292 * frame_num for frames or 2*frame_num for field pics.
293 */
294 int curr_pic_num;
115329f1 295
0da71265
MN
296 /**
297 * max_frame_num or 2*max_frame_num for field pics.
298 */
299 int max_pic_num;
300
301 //Weighted pred stuff
9f2d1b4f
LM
302 int use_weight;
303 int use_weight_chroma;
0da71265
MN
304 int luma_log2_weight_denom;
305 int chroma_log2_weight_denom;
5d18eaad
LM
306 int luma_weight[2][48];
307 int luma_offset[2][48];
308 int chroma_weight[2][48][2];
309 int chroma_offset[2][48][2];
310 int implicit_weight[48][48];
115329f1 311
0da71265 312 //deblock
115329f1 313 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
980a82b7
MN
314 int slice_alpha_c0_offset;
315 int slice_beta_offset;
115329f1 316
0da71265 317 int redundant_pic_count;
115329f1 318
0da71265 319 int direct_spatial_mv_pred;
5ad984c9 320 int dist_scale_factor[16];
5d18eaad 321 int dist_scale_factor_field[32];
2f944356 322 int map_col_to_list0[2][16];
5d18eaad 323 int map_col_to_list0_field[2][32];
0da71265
MN
324
325 /**
326 * num_ref_idx_l0/1_active_minus1 + 1
327 */
88e7a4d1 328 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
187696fa 329 unsigned int list_count;
17107065
LM
330 Picture *short_ref[32];
331 Picture *long_ref[32];
0da71265 332 Picture default_ref_list[2][32];
5d18eaad 333 Picture ref_list[2][48]; ///< 0..15: frame refs, 16..47: mbaff field refs
6f9ca74b 334 Picture *delayed_pic[18]; //FIXME size?
4e4d983e 335 Picture *delayed_output_pic;
115329f1 336
0da71265
MN
337 /**
338 * memory management control operations buffer.
339 */
340 MMCO mmco[MAX_MMCO_COUNT];
341 int mmco_index;
115329f1 342
0da71265
MN
343 int long_ref_count; ///< number of actual long term references
344 int short_ref_count; ///< number of actual short term references
115329f1 345
0da71265
MN
346 //data partitioning
347 GetBitContext intra_gb;
348 GetBitContext inter_gb;
349 GetBitContext *intra_gb_ptr;
350 GetBitContext *inter_gb_ptr;
115329f1 351
68b51e58 352 DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
88e7a4d1 353 DCTELEM mb_padding[256]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not to large or ensure that there is some unused stuff after mb
e5017ab8
LA
354
355 /**
356 * Cabac
357 */
358 CABACContext cabac;
43efd19a 359 uint8_t cabac_state[460];
e5017ab8
LA
360 int cabac_init_idc;
361
362 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
363 uint16_t *cbp_table;
776fe466 364 int cbp;
3981c385
MN
365 int top_cbp;
366 int left_cbp;
693399ba 367 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
e5017ab8
LA
368 uint8_t *chroma_pred_mode_table;
369 int last_qscale_diff;
9e528114 370 int16_t (*mvd_table[2])[2];
68b51e58 371 DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
5ad984c9
LM
372 uint8_t *direct_table;
373 uint8_t direct_cache[5*8];
e5017ab8 374
e4b36d44 375 uint8_t zigzag_scan[16];
548a1c8a
LM
376 uint8_t zigzag_scan8x8[64];
377 uint8_t zigzag_scan8x8_cavlc[64];
5d18eaad
LM
378 uint8_t field_scan[16];
379 uint8_t field_scan8x8[64];
380 uint8_t field_scan8x8_cavlc[64];
36940eca 381 const uint8_t *zigzag_scan_q0;
548a1c8a
LM
382 const uint8_t *zigzag_scan8x8_q0;
383 const uint8_t *zigzag_scan8x8_cavlc_q0;
5d18eaad
LM
384 const uint8_t *field_scan_q0;
385 const uint8_t *field_scan8x8_q0;
386 const uint8_t *field_scan8x8_cavlc_q0;
115329f1 387
cdd10689 388 int x264_build;
0da71265
MN
389}H264Context;
390
391static VLC coeff_token_vlc[4];
392static VLC chroma_dc_coeff_token_vlc;
393
394static VLC total_zeros_vlc[15];
395static VLC chroma_dc_total_zeros_vlc[3];
396
397static VLC run_vlc[6];
398static VLC run7_vlc;
399
8b82a956
MN
400static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
401static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
6ba71fc4 402static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
3e20143e 403static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
8b82a956 404
849f1035 405static av_always_inline uint32_t pack16to32(int a, int b){
377ec888
MN
406#ifdef WORDS_BIGENDIAN
407 return (b&0xFFFF) + (a<<16);
408#else
409 return (a&0xFFFF) + (b<<16);
410#endif
411}
412
acd8d10f
PI
413const uint8_t ff_rem6[52]={
4140, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
415};
416
417const uint8_t ff_div6[52]={
4180, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8,
419};
420
421
0da71265
MN
422/**
423 * fill a rectangle.
5175b937
LLL
424 * @param h height of the rectangle, should be a constant
425 * @param w width of the rectangle, should be a constant
0da71265
MN
426 * @param size the size of val (1 or 4), should be a constant
427 */
849f1035 428static av_always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
af6e2fed 429 uint8_t *p= (uint8_t*)vp;
0da71265 430 assert(size==1 || size==4);
67a82086 431 assert(w<=4);
115329f1 432
0da71265
MN
433 w *= size;
434 stride *= size;
115329f1 435
4733abcb 436 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
d9c780a8 437 assert((stride&(w-1))==0);
67a82086
LM
438 if(w==2){
439 const uint16_t v= size==4 ? val : val*0x0101;
440 *(uint16_t*)(p + 0*stride)= v;
441 if(h==1) return;
442 *(uint16_t*)(p + 1*stride)= v;
443 if(h==2) return;
0da71265 444 *(uint16_t*)(p + 2*stride)=
67a82086
LM
445 *(uint16_t*)(p + 3*stride)= v;
446 }else if(w==4){
447 const uint32_t v= size==4 ? val : val*0x01010101;
448 *(uint32_t*)(p + 0*stride)= v;
449 if(h==1) return;
450 *(uint32_t*)(p + 1*stride)= v;
451 if(h==2) return;
0da71265 452 *(uint32_t*)(p + 2*stride)=
67a82086
LM
453 *(uint32_t*)(p + 3*stride)= v;
454 }else if(w==8){
455 //gcc can't optimize 64bit math on x86_32
456#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
457 const uint64_t v= val*0x0100000001ULL;
458 *(uint64_t*)(p + 0*stride)= v;
459 if(h==1) return;
460 *(uint64_t*)(p + 1*stride)= v;
461 if(h==2) return;
0da71265 462 *(uint64_t*)(p + 2*stride)=
67a82086
LM
463 *(uint64_t*)(p + 3*stride)= v;
464 }else if(w==16){
465 const uint64_t v= val*0x0100000001ULL;
0da71265
MN
466 *(uint64_t*)(p + 0+0*stride)=
467 *(uint64_t*)(p + 8+0*stride)=
468 *(uint64_t*)(p + 0+1*stride)=
67a82086
LM
469 *(uint64_t*)(p + 8+1*stride)= v;
470 if(h==2) return;
0da71265
MN
471 *(uint64_t*)(p + 0+2*stride)=
472 *(uint64_t*)(p + 8+2*stride)=
473 *(uint64_t*)(p + 0+3*stride)=
67a82086
LM
474 *(uint64_t*)(p + 8+3*stride)= v;
475#else
476 *(uint32_t*)(p + 0+0*stride)=
477 *(uint32_t*)(p + 4+0*stride)= val;
478 if(h==1) return;
479 *(uint32_t*)(p + 0+1*stride)=
480 *(uint32_t*)(p + 4+1*stride)= val;
481 if(h==2) return;
482 *(uint32_t*)(p + 0+2*stride)=
483 *(uint32_t*)(p + 4+2*stride)=
484 *(uint32_t*)(p + 0+3*stride)=
485 *(uint32_t*)(p + 4+3*stride)= val;
486 }else if(w==16){
487 *(uint32_t*)(p + 0+0*stride)=
488 *(uint32_t*)(p + 4+0*stride)=
489 *(uint32_t*)(p + 8+0*stride)=
490 *(uint32_t*)(p +12+0*stride)=
491 *(uint32_t*)(p + 0+1*stride)=
492 *(uint32_t*)(p + 4+1*stride)=
493 *(uint32_t*)(p + 8+1*stride)=
494 *(uint32_t*)(p +12+1*stride)= val;
495 if(h==2) return;
496 *(uint32_t*)(p + 0+2*stride)=
497 *(uint32_t*)(p + 4+2*stride)=
498 *(uint32_t*)(p + 8+2*stride)=
499 *(uint32_t*)(p +12+2*stride)=
500 *(uint32_t*)(p + 0+3*stride)=
501 *(uint32_t*)(p + 4+3*stride)=
502 *(uint32_t*)(p + 8+3*stride)=
503 *(uint32_t*)(p +12+3*stride)= val;
504#endif
0da71265
MN
505 }else
506 assert(0);
67a82086 507 assert(h==4);
0da71265
MN
508}
509
70abb407 510static void fill_caches(H264Context *h, int mb_type, int for_deblock){
0da71265 511 MpegEncContext * const s = &h->s;
7bc9090a 512 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
513 int topleft_xy, top_xy, topright_xy, left_xy[2];
514 int topleft_type, top_type, topright_type, left_type[2];
6867a90b 515 int left_block[8];
0da71265
MN
516 int i;
517
717b1733
LM
518 //FIXME deblocking could skip the intra and nnz parts.
519 if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[mb_xy-s->mb_stride]) && !FRAME_MBAFF)
e2e5894a
LM
520 return;
521
115329f1
DB
522 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
523
6867a90b
LLL
524 top_xy = mb_xy - s->mb_stride;
525 topleft_xy = top_xy - 1;
526 topright_xy= top_xy + 1;
527 left_xy[1] = left_xy[0] = mb_xy-1;
528 left_block[0]= 0;
529 left_block[1]= 1;
530 left_block[2]= 2;
531 left_block[3]= 3;
532 left_block[4]= 7;
533 left_block[5]= 10;
534 left_block[6]= 8;
535 left_block[7]= 11;
5d18eaad 536 if(FRAME_MBAFF){
6867a90b
LLL
537 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
538 const int top_pair_xy = pair_xy - s->mb_stride;
539 const int topleft_pair_xy = top_pair_xy - 1;
540 const int topright_pair_xy = top_pair_xy + 1;
541 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
542 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
543 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
544 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
545 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
546 const int bottom = (s->mb_y & 1);
547 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
548 if (bottom
549 ? !curr_mb_frame_flag // bottom macroblock
550 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
551 ) {
552 top_xy -= s->mb_stride;
553 }
554 if (bottom
555 ? !curr_mb_frame_flag // bottom macroblock
556 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
557 ) {
558 topleft_xy -= s->mb_stride;
559 }
560 if (bottom
561 ? !curr_mb_frame_flag // bottom macroblock
562 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
563 ) {
564 topright_xy -= s->mb_stride;
565 }
566 if (left_mb_frame_flag != curr_mb_frame_flag) {
567 left_xy[1] = left_xy[0] = pair_xy - 1;
568 if (curr_mb_frame_flag) {
569 if (bottom) {
570 left_block[0]= 2;
571 left_block[1]= 2;
572 left_block[2]= 3;
573 left_block[3]= 3;
574 left_block[4]= 8;
575 left_block[5]= 11;
576 left_block[6]= 8;
577 left_block[7]= 11;
578 } else {
579 left_block[0]= 0;
580 left_block[1]= 0;
581 left_block[2]= 1;
582 left_block[3]= 1;
583 left_block[4]= 7;
584 left_block[5]= 10;
585 left_block[6]= 7;
586 left_block[7]= 10;
587 }
588 } else {
589 left_xy[1] += s->mb_stride;
590 //left_block[0]= 0;
591 left_block[1]= 2;
592 left_block[2]= 0;
593 left_block[3]= 2;
594 //left_block[4]= 7;
595 left_block[5]= 10;
596 left_block[6]= 7;
597 left_block[7]= 10;
598 }
599 }
0da71265
MN
600 }
601
826de46e
LLL
602 h->top_mb_xy = top_xy;
603 h->left_mb_xy[0] = left_xy[0];
604 h->left_mb_xy[1] = left_xy[1];
6ba71fc4 605 if(for_deblock){
717b1733
LM
606 topleft_type = 0;
607 topright_type = 0;
46f2f05f 608 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
46f2f05f
MN
609 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
610 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
5d18eaad
LM
611
612 if(FRAME_MBAFF && !IS_INTRA(mb_type)){
613 int list;
614 int v = *(uint16_t*)&h->non_zero_count[mb_xy][14];
615 for(i=0; i<16; i++)
616 h->non_zero_count_cache[scan8[i]] = (v>>i)&1;
3425501d 617 for(list=0; list<h->list_count; list++){
5d18eaad
LM
618 if(USES_LIST(mb_type,list)){
619 uint32_t *src = (uint32_t*)s->current_picture.motion_val[list][h->mb2b_xy[mb_xy]];
620 uint32_t *dst = (uint32_t*)h->mv_cache[list][scan8[0]];
191e8ca7 621 int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
5d18eaad
LM
622 for(i=0; i<4; i++, dst+=8, src+=h->b_stride){
623 dst[0] = src[0];
624 dst[1] = src[1];
625 dst[2] = src[2];
626 dst[3] = src[3];
627 }
628 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
629 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = pack16to32(ref[0],ref[1])*0x0101;
630 ref += h->b8_stride;
631 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
632 *(uint32_t*)&h->ref_cache[list][scan8[10]] = pack16to32(ref[0],ref[1])*0x0101;
633 }else{
634 fill_rectangle(&h-> mv_cache[list][scan8[ 0]], 4, 4, 8, 0, 4);
635 fill_rectangle(&h->ref_cache[list][scan8[ 0]], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
636 }
637 }
638 }
46f2f05f
MN
639 }else{
640 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
641 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
642 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
643 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
644 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
645 }
0da71265
MN
646
647 if(IS_INTRA(mb_type)){
115329f1
DB
648 h->topleft_samples_available=
649 h->top_samples_available=
0da71265
MN
650 h->left_samples_available= 0xFFFF;
651 h->topright_samples_available= 0xEEEA;
652
653 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
654 h->topleft_samples_available= 0xB3FF;
655 h->top_samples_available= 0x33FF;
656 h->topright_samples_available= 0x26EA;
657 }
658 for(i=0; i<2; i++){
659 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
660 h->topleft_samples_available&= 0xDF5F;
661 h->left_samples_available&= 0x5F5F;
662 }
663 }
115329f1 664
0da71265
MN
665 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
666 h->topleft_samples_available&= 0x7FFF;
115329f1 667
0da71265
MN
668 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
669 h->topright_samples_available&= 0xFBFF;
115329f1 670
0da71265
MN
671 if(IS_INTRA4x4(mb_type)){
672 if(IS_INTRA4x4(top_type)){
673 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
674 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
675 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
676 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
677 }else{
678 int pred;
6fbcaaa0 679 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
0da71265 680 pred= -1;
6fbcaaa0
LLL
681 else{
682 pred= 2;
0da71265
MN
683 }
684 h->intra4x4_pred_mode_cache[4+8*0]=
685 h->intra4x4_pred_mode_cache[5+8*0]=
686 h->intra4x4_pred_mode_cache[6+8*0]=
687 h->intra4x4_pred_mode_cache[7+8*0]= pred;
688 }
689 for(i=0; i<2; i++){
690 if(IS_INTRA4x4(left_type[i])){
691 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
692 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
693 }else{
694 int pred;
6fbcaaa0 695 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
0da71265 696 pred= -1;
6fbcaaa0
LLL
697 else{
698 pred= 2;
0da71265
MN
699 }
700 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
701 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
702 }
703 }
704 }
705 }
115329f1
DB
706
707
0da71265 708/*
115329f1
DB
7090 . T T. T T T T
7101 L . .L . . . .
7112 L . .L . . . .
7123 . T TL . . . .
7134 L . .L . . . .
7145 L . .. . . . .
0da71265
MN
715*/
716//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
717 if(top_type){
6867a90b
LLL
718 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
719 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
720 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
53c05b1e 721 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
115329f1 722
6867a90b 723 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
53c05b1e 724 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
115329f1 725
6867a90b 726 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
53c05b1e 727 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
115329f1 728
0da71265 729 }else{
115329f1 730 h->non_zero_count_cache[4+8*0]=
0da71265
MN
731 h->non_zero_count_cache[5+8*0]=
732 h->non_zero_count_cache[6+8*0]=
733 h->non_zero_count_cache[7+8*0]=
115329f1 734
0da71265
MN
735 h->non_zero_count_cache[1+8*0]=
736 h->non_zero_count_cache[2+8*0]=
115329f1 737
0da71265 738 h->non_zero_count_cache[1+8*3]=
3981c385 739 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
115329f1 740
0da71265 741 }
826de46e 742
6867a90b
LLL
743 for (i=0; i<2; i++) {
744 if(left_type[i]){
745 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
746 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
747 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
748 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
6867a90b 749 }else{
115329f1
DB
750 h->non_zero_count_cache[3+8*1 + 2*8*i]=
751 h->non_zero_count_cache[3+8*2 + 2*8*i]=
752 h->non_zero_count_cache[0+8*1 + 8*i]=
6867a90b 753 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
826de46e
LLL
754 }
755 }
756
757 if( h->pps.cabac ) {
758 // top_cbp
759 if(top_type) {
760 h->top_cbp = h->cbp_table[top_xy];
761 } else if(IS_INTRA(mb_type)) {
762 h->top_cbp = 0x1C0;
763 } else {
764 h->top_cbp = 0;
765 }
766 // left_cbp
767 if (left_type[0]) {
768 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
769 } else if(IS_INTRA(mb_type)) {
770 h->left_cbp = 0x1C0;
771 } else {
772 h->left_cbp = 0;
773 }
774 if (left_type[0]) {
775 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
776 }
777 if (left_type[1]) {
778 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
6867a90b 779 }
0da71265 780 }
6867a90b 781
0da71265 782#if 1
e2e5894a 783 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
0da71265 784 int list;
3425501d 785 for(list=0; list<h->list_count; list++){
e2e5894a 786 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
0da71265
MN
787 /*if(!h->mv_cache_clean[list]){
788 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
789 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
790 h->mv_cache_clean[list]= 1;
791 }*/
5ad984c9 792 continue;
0da71265
MN
793 }
794 h->mv_cache_clean[list]= 0;
115329f1 795
53b19144 796 if(USES_LIST(top_type, list)){
0da71265
MN
797 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
798 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
799 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
800 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
801 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
802 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
803 h->ref_cache[list][scan8[0] + 0 - 1*8]=
804 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
805 h->ref_cache[list][scan8[0] + 2 - 1*8]=
806 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
807 }else{
115329f1
DB
808 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
809 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
810 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
0da71265
MN
811 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
812 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
813 }
814
4672503d
LM
815 for(i=0; i<2; i++){
816 int cache_idx = scan8[0] - 1 + i*2*8;
817 if(USES_LIST(left_type[i], list)){
818 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
819 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
820 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
821 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
822 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
823 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
824 }else{
825 *(uint32_t*)h->mv_cache [list][cache_idx ]=
826 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
827 h->ref_cache[list][cache_idx ]=
828 h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
829 }
0da71265
MN
830 }
831
ae08a563 832 if((for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred)) && !FRAME_MBAFF)
46f2f05f
MN
833 continue;
834
53b19144 835 if(USES_LIST(topleft_type, list)){
e2e5894a
LM
836 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
837 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
838 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
839 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
840 }else{
841 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
842 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
843 }
115329f1 844
53b19144 845 if(USES_LIST(topright_type, list)){
e2e5894a
LM
846 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
847 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
848 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
849 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
850 }else{
851 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
852 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
853 }
e2e5894a 854
ae08a563 855 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
717b1733 856 continue;
115329f1
DB
857
858 h->ref_cache[list][scan8[5 ]+1] =
859 h->ref_cache[list][scan8[7 ]+1] =
3b66c4c5 860 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
115329f1 861 h->ref_cache[list][scan8[4 ]] =
0da71265
MN
862 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
863 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
864 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
3b66c4c5 865 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
0da71265
MN
866 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
867 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
9e528114
LA
868
869 if( h->pps.cabac ) {
870 /* XXX beurk, Load mvd */
53b19144 871 if(USES_LIST(top_type, list)){
9e528114
LA
872 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
873 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
874 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
875 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
876 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
877 }else{
115329f1
DB
878 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
879 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
880 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
9e528114
LA
881 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
882 }
53b19144 883 if(USES_LIST(left_type[0], list)){
9e528114
LA
884 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
885 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
886 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
887 }else{
888 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
889 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
890 }
53b19144 891 if(USES_LIST(left_type[1], list)){
9e528114
LA
892 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
893 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
894 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
895 }else{
896 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
897 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
898 }
899 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
900 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
3b66c4c5 901 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
9e528114
LA
902 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
903 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
5ad984c9
LM
904
905 if(h->slice_type == B_TYPE){
906 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
907
908 if(IS_DIRECT(top_type)){
909 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
910 }else if(IS_8X8(top_type)){
911 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
912 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
913 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
914 }else{
915 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
916 }
115329f1 917
5d18eaad
LM
918 if(IS_DIRECT(left_type[0]))
919 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
920 else if(IS_8X8(left_type[0]))
921 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
922 else
923 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
924
925 if(IS_DIRECT(left_type[1]))
5ad984c9 926 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
5d18eaad
LM
927 else if(IS_8X8(left_type[1]))
928 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
929 else
5ad984c9 930 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
5d18eaad
LM
931 }
932 }
933
934 if(FRAME_MBAFF){
935#define MAP_MVS\
936 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
937 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
938 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
939 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
940 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
941 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
942 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
943 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
944 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
945 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
946 if(MB_FIELD){
947#define MAP_F2F(idx, mb_type)\
948 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
949 h->ref_cache[list][idx] <<= 1;\
950 h->mv_cache[list][idx][1] /= 2;\
951 h->mvd_cache[list][idx][1] /= 2;\
952 }
953 MAP_MVS
954#undef MAP_F2F
955 }else{
956#define MAP_F2F(idx, mb_type)\
957 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
958 h->ref_cache[list][idx] >>= 1;\
959 h->mv_cache[list][idx][1] <<= 1;\
960 h->mvd_cache[list][idx][1] <<= 1;\
5ad984c9 961 }
5d18eaad
LM
962 MAP_MVS
963#undef MAP_F2F
5ad984c9 964 }
9e528114 965 }
0da71265 966 }
0da71265
MN
967 }
968#endif
43efd19a
LM
969
970 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
0da71265
MN
971}
972
973static inline void write_back_intra_pred_mode(H264Context *h){
974 MpegEncContext * const s = &h->s;
7bc9090a 975 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
976
977 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
978 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
979 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
980 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
981 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
982 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
983 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
984}
985
986/**
987 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
988 */
989static inline int check_intra4x4_pred_mode(H264Context *h){
990 MpegEncContext * const s = &h->s;
991 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
992 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
993 int i;
115329f1 994
0da71265
MN
995 if(!(h->top_samples_available&0x8000)){
996 for(i=0; i<4; i++){
997 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
998 if(status<0){
9b879566 999 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);
0da71265
MN
1000 return -1;
1001 } else if(status){
1002 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
1003 }
1004 }
1005 }
115329f1 1006
0da71265
MN
1007 if(!(h->left_samples_available&0x8000)){
1008 for(i=0; i<4; i++){
1009 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
1010 if(status<0){
9b879566 1011 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);
0da71265
MN
1012 return -1;
1013 } else if(status){
1014 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
1015 }
1016 }
1017 }
1018
1019 return 0;
1020} //FIXME cleanup like next
1021
1022/**
1023 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
1024 */
1025static inline int check_intra_pred_mode(H264Context *h, int mode){
1026 MpegEncContext * const s = &h->s;
1027 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
1028 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 1029
43ff0714 1030 if(mode > 6U) {
5175b937 1031 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
7440fe83 1032 return -1;
5175b937 1033 }
115329f1 1034
0da71265
MN
1035 if(!(h->top_samples_available&0x8000)){
1036 mode= top[ mode ];
1037 if(mode<0){
9b879566 1038 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
1039 return -1;
1040 }
1041 }
115329f1 1042
0da71265
MN
1043 if(!(h->left_samples_available&0x8000)){
1044 mode= left[ mode ];
1045 if(mode<0){
9b879566 1046 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265 1047 return -1;
115329f1 1048 }
0da71265
MN
1049 }
1050
1051 return mode;
1052}
1053
1054/**
1055 * gets the predicted intra4x4 prediction mode.
1056 */
1057static inline int pred_intra_mode(H264Context *h, int n){
1058 const int index8= scan8[n];
1059 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1060 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1061 const int min= FFMIN(left, top);
1062
95c26348 1063 tprintf("mode:%d %d min:%d\n", left ,top, min);
0da71265
MN
1064
1065 if(min<0) return DC_PRED;
1066 else return min;
1067}
1068
1069static inline void write_back_non_zero_count(H264Context *h){
1070 MpegEncContext * const s = &h->s;
7bc9090a 1071 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 1072
6867a90b
LLL
1073 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1074 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1075 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
53c05b1e 1076 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
6867a90b
LLL
1077 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1078 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1079 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
115329f1 1080
6867a90b 1081 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
53c05b1e 1082 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
6867a90b 1083 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
53c05b1e 1084
6867a90b 1085 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
53c05b1e 1086 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
6867a90b 1087 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
5d18eaad
LM
1088
1089 if(FRAME_MBAFF){
1090 // store all luma nnzs, for deblocking
1091 int v = 0, i;
1092 for(i=0; i<16; i++)
1093 v += (!!h->non_zero_count_cache[scan8[i]]) << i;
1094 *(uint16_t*)&h->non_zero_count[mb_xy][14] = v;
1095 }
0da71265
MN
1096}
1097
1098/**
1099 * gets the predicted number of non zero coefficients.
1100 * @param n block index
1101 */
1102static inline int pred_non_zero_count(H264Context *h, int n){
1103 const int index8= scan8[n];
1104 const int left= h->non_zero_count_cache[index8 - 1];
1105 const int top = h->non_zero_count_cache[index8 - 8];
1106 int i= left + top;
115329f1 1107
0da71265
MN
1108 if(i<64) i= (i+1)>>1;
1109
95c26348 1110 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
0da71265
MN
1111
1112 return i&31;
1113}
1114
1924f3ce
MN
1115static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1116 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1117
5d18eaad
LM
1118 /* there is no consistent mapping of mvs to neighboring locations that will
1119 * make mbaff happy, so we can't move all this logic to fill_caches */
1120 if(FRAME_MBAFF){
1121 MpegEncContext *s = &h->s;
191e8ca7 1122 const uint32_t *mb_types = s->current_picture_ptr->mb_type;
5d18eaad
LM
1123 const int16_t *mv;
1124 *(uint32_t*)h->mv_cache[list][scan8[0]-2] = 0;
1125 *C = h->mv_cache[list][scan8[0]-2];
1126
1127 if(!MB_FIELD
1128 && (s->mb_y&1) && i < scan8[0]+8 && topright_ref != PART_NOT_AVAILABLE){
1129 int topright_xy = s->mb_x + (s->mb_y-1)*s->mb_stride + (i == scan8[0]+3);
1130 if(IS_INTERLACED(mb_types[topright_xy])){
1131#define SET_DIAG_MV(MV_OP, REF_OP, X4, Y4)\
1132 const int x4 = X4, y4 = Y4;\
1133 const int mb_type = mb_types[(x4>>2)+(y4>>2)*s->mb_stride];\
1134 if(!USES_LIST(mb_type,list) && !IS_8X8(mb_type))\
1135 return LIST_NOT_USED;\
1136 mv = s->current_picture_ptr->motion_val[list][x4 + y4*h->b_stride];\
1137 h->mv_cache[list][scan8[0]-2][0] = mv[0];\
1138 h->mv_cache[list][scan8[0]-2][1] = mv[1] MV_OP;\
1139 return s->current_picture_ptr->ref_index[list][(x4>>1) + (y4>>1)*h->b8_stride] REF_OP;
1140
1141 SET_DIAG_MV(*2, >>1, s->mb_x*4+(i&7)-4+part_width, s->mb_y*4-1);
1142 }
1143 }
1144 if(topright_ref == PART_NOT_AVAILABLE
1145 && ((s->mb_y&1) || i >= scan8[0]+8) && (i&7)==4
1146 && h->ref_cache[list][scan8[0]-1] != PART_NOT_AVAILABLE){
1147 if(!MB_FIELD
1148 && IS_INTERLACED(mb_types[h->left_mb_xy[0]])){
1149 SET_DIAG_MV(*2, >>1, s->mb_x*4-1, (s->mb_y|1)*4+(s->mb_y&1)*2+(i>>4)-1);
1150 }
1151 if(MB_FIELD
1152 && !IS_INTERLACED(mb_types[h->left_mb_xy[0]])
1153 && i >= scan8[0]+8){
1154 // leftshift will turn LIST_NOT_USED into PART_NOT_AVAILABLE, but that's ok.
1155 SET_DIAG_MV(>>1, <<1, s->mb_x*4-1, (s->mb_y&~1)*4 - 1 + ((i-scan8[0])>>3)*2);
1156 }
1157 }
1158#undef SET_DIAG_MV
1159 }
1160
1924f3ce
MN
1161 if(topright_ref != PART_NOT_AVAILABLE){
1162 *C= h->mv_cache[list][ i - 8 + part_width ];
1163 return topright_ref;
1164 }else{
95c26348
MN
1165 tprintf("topright MV not available\n");
1166
1924f3ce
MN
1167 *C= h->mv_cache[list][ i - 8 - 1 ];
1168 return h->ref_cache[list][ i - 8 - 1 ];
1169 }
1170}
1171
0da71265
MN
1172/**
1173 * gets the predicted MV.
1174 * @param n the block index
1175 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1176 * @param mx the x component of the predicted motion vector
1177 * @param my the y component of the predicted motion vector
1178 */
1179static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
0da71265
MN
1180 const int index8= scan8[n];
1181 const int top_ref= h->ref_cache[list][ index8 - 8 ];
0da71265
MN
1182 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1183 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1184 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1924f3ce
MN
1185 const int16_t * C;
1186 int diagonal_ref, match_count;
1187
0da71265 1188 assert(part_width==1 || part_width==2 || part_width==4);
1924f3ce 1189
0da71265 1190/* mv_cache
115329f1 1191 B . . A T T T T
0da71265
MN
1192 U . . L . . , .
1193 U . . L . . . .
1194 U . . L . . , .
1195 . . . L . . . .
1196*/
1924f3ce
MN
1197
1198 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1199 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
827c91bf 1200 tprintf("pred_motion match_count=%d\n", match_count);
1924f3ce
MN
1201 if(match_count > 1){ //most common
1202 *mx= mid_pred(A[0], B[0], C[0]);
1203 *my= mid_pred(A[1], B[1], C[1]);
1204 }else if(match_count==1){
1205 if(left_ref==ref){
1206 *mx= A[0];
115329f1 1207 *my= A[1];
1924f3ce
MN
1208 }else if(top_ref==ref){
1209 *mx= B[0];
115329f1 1210 *my= B[1];
0da71265 1211 }else{
1924f3ce 1212 *mx= C[0];
115329f1 1213 *my= C[1];
0da71265
MN
1214 }
1215 }else{
1924f3ce 1216 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
0da71265 1217 *mx= A[0];
115329f1 1218 *my= A[1];
0da71265 1219 }else{
1924f3ce
MN
1220 *mx= mid_pred(A[0], B[0], C[0]);
1221 *my= mid_pred(A[1], B[1], C[1]);
0da71265 1222 }
0da71265 1223 }
115329f1 1224
af6e2fed 1225 tprintf("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);
0da71265
MN
1226}
1227
1228/**
1229 * gets the directionally predicted 16x8 MV.
1230 * @param n the block index
1231 * @param mx the x component of the predicted motion vector
1232 * @param my the y component of the predicted motion vector
1233 */
1234static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1235 if(n==0){
1236 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1237 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1238
579448a0 1239 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
115329f1 1240
0da71265
MN
1241 if(top_ref == ref){
1242 *mx= B[0];
1243 *my= B[1];
1244 return;
1245 }
1246 }else{
1247 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1248 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
115329f1 1249
579448a0 1250 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1251
1252 if(left_ref == ref){
1253 *mx= A[0];
1254 *my= A[1];
1255 return;
1256 }
1257 }
1258
1259 //RARE
1260 pred_motion(h, n, 4, list, ref, mx, my);
1261}
1262
1263/**
1264 * gets the directionally predicted 8x16 MV.
1265 * @param n the block index
1266 * @param mx the x component of the predicted motion vector
1267 * @param my the y component of the predicted motion vector
1268 */
1269static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1270 if(n==0){
1271 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1272 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
115329f1 1273
579448a0 1274 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1275
1276 if(left_ref == ref){
1277 *mx= A[0];
1278 *my= A[1];
1279 return;
1280 }
1281 }else{
1924f3ce
MN
1282 const int16_t * C;
1283 int diagonal_ref;
1284
1285 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
115329f1 1286
579448a0 1287 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265 1288
115329f1 1289 if(diagonal_ref == ref){
0da71265
MN
1290 *mx= C[0];
1291 *my= C[1];
1292 return;
1293 }
0da71265
MN
1294 }
1295
1296 //RARE
1297 pred_motion(h, n, 2, list, ref, mx, my);
1298}
1299
1300static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
0da71265
MN
1301 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1302 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1303
5175b937 1304 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
0da71265
MN
1305
1306 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1307 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1308 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
115329f1 1309
0da71265
MN
1310 *mx = *my = 0;
1311 return;
1312 }
115329f1 1313
0da71265
MN
1314 pred_motion(h, 0, 4, 0, 0, mx, my);
1315
1316 return;
1317}
1318
5ad984c9
LM
1319static inline void direct_dist_scale_factor(H264Context * const h){
1320 const int poc = h->s.current_picture_ptr->poc;
1321 const int poc1 = h->ref_list[1][0].poc;
1322 int i;
1323 for(i=0; i<h->ref_count[0]; i++){
1324 int poc0 = h->ref_list[0][i].poc;
1325 int td = clip(poc1 - poc0, -128, 127);
1326 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1327 h->dist_scale_factor[i] = 256;
1328 }else{
1329 int tb = clip(poc - poc0, -128, 127);
c26abfa5 1330 int tx = (16384 + (FFABS(td) >> 1)) / td;
5ad984c9
LM
1331 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1332 }
1333 }
5d18eaad
LM
1334 if(FRAME_MBAFF){
1335 for(i=0; i<h->ref_count[0]; i++){
1336 h->dist_scale_factor_field[2*i] =
1337 h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
1338 }
1339 }
5ad984c9 1340}
2f944356
LM
1341static inline void direct_ref_list_init(H264Context * const h){
1342 MpegEncContext * const s = &h->s;
1343 Picture * const ref1 = &h->ref_list[1][0];
1344 Picture * const cur = s->current_picture_ptr;
1345 int list, i, j;
1346 if(cur->pict_type == I_TYPE)
1347 cur->ref_count[0] = 0;
1348 if(cur->pict_type != B_TYPE)
1349 cur->ref_count[1] = 0;
1350 for(list=0; list<2; list++){
1351 cur->ref_count[list] = h->ref_count[list];
1352 for(j=0; j<h->ref_count[list]; j++)
1353 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1354 }
1355 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1356 return;
1357 for(list=0; list<2; list++){
1358 for(i=0; i<ref1->ref_count[list]; i++){
1359 const int poc = ref1->ref_poc[list][i];
171c4076 1360 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
2f944356
LM
1361 for(j=0; j<h->ref_count[list]; j++)
1362 if(h->ref_list[list][j].poc == poc){
1363 h->map_col_to_list0[list][i] = j;
1364 break;
1365 }
1366 }
1367 }
5d18eaad
LM
1368 if(FRAME_MBAFF){
1369 for(list=0; list<2; list++){
1370 for(i=0; i<ref1->ref_count[list]; i++){
1371 j = h->map_col_to_list0[list][i];
1372 h->map_col_to_list0_field[list][2*i] = 2*j;
1373 h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
1374 }
1375 }
1376 }
2f944356 1377}
5ad984c9
LM
1378
1379static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1380 MpegEncContext * const s = &h->s;
1381 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1382 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1383 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1384 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1385 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
4866bd2b 1386 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
5ad984c9 1387 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
2f944356 1388 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
5ad984c9 1389 const int is_b8x8 = IS_8X8(*mb_type);
88e7a4d1 1390 unsigned int sub_mb_type;
5ad984c9
LM
1391 int i8, i4;
1392
5d18eaad 1393#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
5ad984c9
LM
1394 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1395 /* FIXME save sub mb types from previous frames (or derive from MVs)
1396 * so we know exactly what block size to use */
1397 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
3622988f 1398 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5d18eaad 1399 }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
5ad984c9
LM
1400 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1401 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1402 }else{
1403 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
3622988f 1404 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1405 }
1406 if(!is_b8x8)
1407 *mb_type |= MB_TYPE_DIRECT2;
5d18eaad
LM
1408 if(MB_FIELD)
1409 *mb_type |= MB_TYPE_INTERLACED;
5ad984c9 1410
827c91bf 1411 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
115329f1 1412
5ad984c9
LM
1413 if(h->direct_spatial_mv_pred){
1414 int ref[2];
1415 int mv[2][2];
1416 int list;
1417
5d18eaad
LM
1418 /* FIXME interlacing + spatial direct uses wrong colocated block positions */
1419
5ad984c9
LM
1420 /* ref = min(neighbors) */
1421 for(list=0; list<2; list++){
1422 int refa = h->ref_cache[list][scan8[0] - 1];
1423 int refb = h->ref_cache[list][scan8[0] - 8];
1424 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1425 if(refc == -2)
1426 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1427 ref[list] = refa;
1428 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1429 ref[list] = refb;
1430 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1431 ref[list] = refc;
1432 if(ref[list] < 0)
1433 ref[list] = -1;
1434 }
1435
1436 if(ref[0] < 0 && ref[1] < 0){
1437 ref[0] = ref[1] = 0;
1438 mv[0][0] = mv[0][1] =
1439 mv[1][0] = mv[1][1] = 0;
1440 }else{
1441 for(list=0; list<2; list++){
1442 if(ref[list] >= 0)
1443 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1444 else
1445 mv[list][0] = mv[list][1] = 0;
1446 }
1447 }
1448
1449 if(ref[1] < 0){
1450 *mb_type &= ~MB_TYPE_P0L1;
1451 sub_mb_type &= ~MB_TYPE_P0L1;
1452 }else if(ref[0] < 0){
1453 *mb_type &= ~MB_TYPE_P0L0;
1454 sub_mb_type &= ~MB_TYPE_P0L0;
1455 }
1456
1457 if(IS_16X16(*mb_type)){
d19f5acb
MN
1458 int a=0, b=0;
1459
cec93959
LM
1460 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1461 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
115329f1 1462 if(!IS_INTRA(mb_type_col)
c26abfa5
DB
1463 && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
1464 || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
bf4e3bd2 1465 && (h->x264_build>33 || !h->x264_build)))){
5ad984c9 1466 if(ref[0] > 0)
d19f5acb 1467 a= pack16to32(mv[0][0],mv[0][1]);
5ad984c9 1468 if(ref[1] > 0)
d19f5acb 1469 b= pack16to32(mv[1][0],mv[1][1]);
5ad984c9 1470 }else{
d19f5acb
MN
1471 a= pack16to32(mv[0][0],mv[0][1]);
1472 b= pack16to32(mv[1][0],mv[1][1]);
5ad984c9 1473 }
d19f5acb
MN
1474 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
1475 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
5ad984c9
LM
1476 }else{
1477 for(i8=0; i8<4; i8++){
1478 const int x8 = i8&1;
1479 const int y8 = i8>>1;
115329f1 1480
5ad984c9
LM
1481 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1482 continue;
1483 h->sub_mb_type[i8] = sub_mb_type;
115329f1 1484
5ad984c9
LM
1485 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1486 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
cec93959
LM
1487 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1488 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
115329f1 1489
5ad984c9 1490 /* col_zero_flag */
115329f1
DB
1491 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1492 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
bf4e3bd2 1493 && (h->x264_build>33 || !h->x264_build)))){
4866bd2b 1494 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
f1f17e54
LM
1495 if(IS_SUB_8X8(sub_mb_type)){
1496 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
c26abfa5 1497 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
f1f17e54
LM
1498 if(ref[0] == 0)
1499 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1500 if(ref[1] == 0)
1501 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1502 }
1503 }else
5ad984c9 1504 for(i4=0; i4<4; i4++){
4866bd2b 1505 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
c26abfa5 1506 if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
5ad984c9
LM
1507 if(ref[0] == 0)
1508 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1509 if(ref[1] == 0)
1510 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1511 }
1512 }
1513 }
1514 }
1515 }
1516 }else{ /* direct temporal mv pred */
5d18eaad
LM
1517 const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
1518 const int *dist_scale_factor = h->dist_scale_factor;
1519
1520 if(FRAME_MBAFF){
1521 if(IS_INTERLACED(*mb_type)){
1522 map_col_to_list0[0] = h->map_col_to_list0_field[0];
1523 map_col_to_list0[1] = h->map_col_to_list0_field[1];
1524 dist_scale_factor = h->dist_scale_factor_field;
1525 }
1526 if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
1527 /* FIXME assumes direct_8x8_inference == 1 */
1528 const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
1529 int mb_types_col[2];
1530 int y_shift;
1531
1532 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
1533 | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
1534 | (*mb_type & MB_TYPE_INTERLACED);
1535 sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;
1536
1537 if(IS_INTERLACED(*mb_type)){
1538 /* frame to field scaling */
1539 mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
1540 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1541 if(s->mb_y&1){
1542 l1ref0 -= 2*h->b8_stride;
1543 l1ref1 -= 2*h->b8_stride;
1544 l1mv0 -= 4*h->b_stride;
1545 l1mv1 -= 4*h->b_stride;
1546 }
1547 y_shift = 0;
1548
1549 if( (mb_types_col[0] & MB_TYPE_16x16_OR_INTRA)
1550 && (mb_types_col[1] & MB_TYPE_16x16_OR_INTRA)
1551 && !is_b8x8)
1552 *mb_type |= MB_TYPE_16x8;
1553 else
1554 *mb_type |= MB_TYPE_8x8;
1555 }else{
1556 /* field to frame scaling */
1557 /* col_mb_y = (mb_y&~1) + (topAbsDiffPOC < bottomAbsDiffPOC ? 0 : 1)
1558 * but in MBAFF, top and bottom POC are equal */
1559 int dy = (s->mb_y&1) ? 1 : 2;
1560 mb_types_col[0] =
1561 mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
1562 l1ref0 += dy*h->b8_stride;
1563 l1ref1 += dy*h->b8_stride;
1564 l1mv0 += 2*dy*h->b_stride;
1565 l1mv1 += 2*dy*h->b_stride;
1566 y_shift = 2;
1567
1568 if((mb_types_col[0] & (MB_TYPE_16x16_OR_INTRA|MB_TYPE_16x8))
1569 && !is_b8x8)
1570 *mb_type |= MB_TYPE_16x16;
1571 else
1572 *mb_type |= MB_TYPE_8x8;
1573 }
1574
1575 for(i8=0; i8<4; i8++){
1576 const int x8 = i8&1;
1577 const int y8 = i8>>1;
1578 int ref0, scale;
1579 const int16_t (*l1mv)[2]= l1mv0;
1580
1581 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1582 continue;
1583 h->sub_mb_type[i8] = sub_mb_type;
1584
1585 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1586 if(IS_INTRA(mb_types_col[y8])){
1587 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1588 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1589 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1590 continue;
1591 }
1592
1593 ref0 = l1ref0[x8 + (y8*2>>y_shift)*h->b8_stride];
1594 if(ref0 >= 0)
1595 ref0 = map_col_to_list0[0][ref0*2>>y_shift];
1596 else{
1597 ref0 = map_col_to_list0[1][l1ref1[x8 + (y8*2>>y_shift)*h->b8_stride]*2>>y_shift];
1598 l1mv= l1mv1;
1599 }
1600 scale = dist_scale_factor[ref0];
1601 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1602
1603 {
1604 const int16_t *mv_col = l1mv[x8*3 + (y8*6>>y_shift)*h->b_stride];
1605 int my_col = (mv_col[1]<<y_shift)/2;
1606 int mx = (scale * mv_col[0] + 128) >> 8;
1607 int my = (scale * my_col + 128) >> 8;
1608 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1609 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
1610 }
1611 }
1612 return;
1613 }
1614 }
1615
1616 /* one-to-one mv scaling */
1617
5ad984c9 1618 if(IS_16X16(*mb_type)){
fda51641
MN
1619 int ref, mv0, mv1;
1620
5ad984c9
LM
1621 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1622 if(IS_INTRA(mb_type_col)){
fda51641 1623 ref=mv0=mv1=0;
5ad984c9 1624 }else{
5d18eaad
LM
1625 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0]]
1626 : map_col_to_list0[1][l1ref1[0]];
1627 const int scale = dist_scale_factor[ref0];
8583bef8 1628 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
5ad984c9 1629 int mv_l0[2];
5d18eaad
LM
1630 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1631 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
fda51641
MN
1632 ref= ref0;
1633 mv0= pack16to32(mv_l0[0],mv_l0[1]);
1634 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
5ad984c9 1635 }
fda51641
MN
1636 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
1637 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
1638 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
5ad984c9
LM
1639 }else{
1640 for(i8=0; i8<4; i8++){
1641 const int x8 = i8&1;
1642 const int y8 = i8>>1;
5d18eaad 1643 int ref0, scale;
bf4e3bd2 1644 const int16_t (*l1mv)[2]= l1mv0;
8583bef8 1645
5ad984c9
LM
1646 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1647 continue;
1648 h->sub_mb_type[i8] = sub_mb_type;
5d18eaad 1649 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
5ad984c9
LM
1650 if(IS_INTRA(mb_type_col)){
1651 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
5ad984c9
LM
1652 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1653 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1654 continue;
1655 }
115329f1 1656
5ad984c9 1657 ref0 = l1ref0[x8 + y8*h->b8_stride];
2f944356 1658 if(ref0 >= 0)
5d18eaad 1659 ref0 = map_col_to_list0[0][ref0];
8583bef8 1660 else{
5d18eaad 1661 ref0 = map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
8583bef8
MN
1662 l1mv= l1mv1;
1663 }
5d18eaad 1664 scale = dist_scale_factor[ref0];
115329f1 1665
5ad984c9 1666 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
f1f17e54
LM
1667 if(IS_SUB_8X8(sub_mb_type)){
1668 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
5d18eaad
LM
1669 int mx = (scale * mv_col[0] + 128) >> 8;
1670 int my = (scale * mv_col[1] + 128) >> 8;
f1f17e54
LM
1671 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1672 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1673 }else
5ad984c9 1674 for(i4=0; i4<4; i4++){
8583bef8 1675 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9 1676 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
5d18eaad
LM
1677 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
1678 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
5ad984c9
LM
1679 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1680 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1681 }
1682 }
1683 }
1684 }
1685}
1686
0da71265
MN
1687static inline void write_back_motion(H264Context *h, int mb_type){
1688 MpegEncContext * const s = &h->s;
0da71265
MN
1689 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1690 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1691 int list;
1692
2ea39252
LM
1693 if(!USES_LIST(mb_type, 0))
1694 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1695
3425501d 1696 for(list=0; list<h->list_count; list++){
0da71265 1697 int y;
53b19144 1698 if(!USES_LIST(mb_type, list))
5ad984c9 1699 continue;
115329f1 1700
0da71265
MN
1701 for(y=0; y<4; y++){
1702 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1703 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1704 }
9e528114 1705 if( h->pps.cabac ) {
e6e77eb6
LM
1706 if(IS_SKIP(mb_type))
1707 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1708 else
9e528114
LA
1709 for(y=0; y<4; y++){
1710 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1711 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1712 }
1713 }
53b19144
LM
1714
1715 {
191e8ca7 1716 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
53b19144
LM
1717 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1718 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1719 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1720 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
0da71265
MN
1721 }
1722 }
115329f1 1723
5ad984c9
LM
1724 if(h->slice_type == B_TYPE && h->pps.cabac){
1725 if(IS_8X8(mb_type)){
53b19144
LM
1726 uint8_t *direct_table = &h->direct_table[b8_xy];
1727 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1728 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1729 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
5ad984c9
LM
1730 }
1731 }
0da71265
MN
1732}
1733
1734/**
1735 * Decodes a network abstraction layer unit.
1736 * @param consumed is the number of bytes used as input
1737 * @param length is the length of the array
3b66c4c5 1738 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
115329f1 1739 * @returns decoded bytes, might be src+1 if no escapes
0da71265
MN
1740 */
1741static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1742 int i, si, di;
1743 uint8_t *dst;
1744
bb270c08 1745// src[0]&0x80; //forbidden bit
0da71265
MN
1746 h->nal_ref_idc= src[0]>>5;
1747 h->nal_unit_type= src[0]&0x1F;
1748
1749 src++; length--;
115329f1 1750#if 0
0da71265
MN
1751 for(i=0; i<length; i++)
1752 printf("%2X ", src[i]);
1753#endif
1754 for(i=0; i+1<length; i+=2){
1755 if(src[i]) continue;
1756 if(i>0 && src[i-1]==0) i--;
1757 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1758 if(src[i+2]!=3){
1759 /* startcode, so we must be past the end */
1760 length=i;
1761 }
1762 break;
1763 }
1764 }
1765
1766 if(i>=length-1){ //no escaped 0
1767 *dst_length= length;
1768 *consumed= length+1; //+1 for the header
115329f1 1769 return src;
0da71265
MN
1770 }
1771
1772 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1773 dst= h->rbsp_buffer;
1774
ac658be5
FOL
1775 if (dst == NULL){
1776 return NULL;
1777 }
1778
3b66c4c5 1779//printf("decoding esc\n");
0da71265 1780 si=di=0;
115329f1 1781 while(si<length){
0da71265
MN
1782 //remove escapes (very rare 1:2^22)
1783 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1784 if(src[si+2]==3){ //escape
1785 dst[di++]= 0;
1786 dst[di++]= 0;
1787 si+=3;
c8470cc1 1788 continue;
0da71265
MN
1789 }else //next start code
1790 break;
1791 }
1792
1793 dst[di++]= src[si++];
1794 }
1795
1796 *dst_length= di;
1797 *consumed= si + 1;//+1 for the header
1798//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1799 return dst;
1800}
1801
0da71265
MN
1802/**
1803 * identifies the exact end of the bitstream
1804 * @return the length of the trailing, or 0 if damaged
1805 */
1806static int decode_rbsp_trailing(uint8_t *src){
1807 int v= *src;
1808 int r;
1809
95c26348 1810 tprintf("rbsp trailing %X\n", v);
0da71265
MN
1811
1812 for(r=1; r<9; r++){
1813 if(v&1) return r;
1814 v>>=1;
1815 }
1816 return 0;
1817}
1818
1819/**
1820 * idct tranforms the 16 dc values and dequantize them.
1821 * @param qp quantization parameter
1822 */
239ea04c 1823static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1824#define stride 16
1825 int i;
1826 int temp[16]; //FIXME check if this is a good idea
1827 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1828 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1829
1830//memset(block, 64, 2*256);
1831//return;
1832 for(i=0; i<4; i++){
1833 const int offset= y_offset[i];
1834 const int z0= block[offset+stride*0] + block[offset+stride*4];
1835 const int z1= block[offset+stride*0] - block[offset+stride*4];
1836 const int z2= block[offset+stride*1] - block[offset+stride*5];
1837 const int z3= block[offset+stride*1] + block[offset+stride*5];
1838
1839 temp[4*i+0]= z0+z3;
1840 temp[4*i+1]= z1+z2;
1841 temp[4*i+2]= z1-z2;
1842 temp[4*i+3]= z0-z3;
1843 }
1844
1845 for(i=0; i<4; i++){
1846 const int offset= x_offset[i];
1847 const int z0= temp[4*0+i] + temp[4*2+i];
1848 const int z1= temp[4*0+i] - temp[4*2+i];
1849 const int z2= temp[4*1+i] - temp[4*3+i];
1850 const int z3= temp[4*1+i] + temp[4*3+i];
1851
239ea04c
LM
1852 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1853 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1854 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1855 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
0da71265
MN
1856 }
1857}
1858
e5017ab8 1859#if 0
0da71265
MN
1860/**
1861 * dct tranforms the 16 dc values.
1862 * @param qp quantization parameter ??? FIXME
1863 */
1864static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1865// const int qmul= dequant_coeff[qp][0];
1866 int i;
1867 int temp[16]; //FIXME check if this is a good idea
1868 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1869 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1870
1871 for(i=0; i<4; i++){
1872 const int offset= y_offset[i];
1873 const int z0= block[offset+stride*0] + block[offset+stride*4];
1874 const int z1= block[offset+stride*0] - block[offset+stride*4];
1875 const int z2= block[offset+stride*1] - block[offset+stride*5];
1876 const int z3= block[offset+stride*1] + block[offset+stride*5];
1877
1878 temp[4*i+0]= z0+z3;
1879 temp[4*i+1]= z1+z2;
1880 temp[4*i+2]= z1-z2;
1881 temp[4*i+3]= z0-z3;
1882 }
1883
1884 for(i=0; i<4; i++){
1885 const int offset= x_offset[i];
1886 const int z0= temp[4*0+i] + temp[4*2+i];
1887 const int z1= temp[4*0+i] - temp[4*2+i];
1888 const int z2= temp[4*1+i] - temp[4*3+i];
1889 const int z3= temp[4*1+i] + temp[4*3+i];
1890
1891 block[stride*0 +offset]= (z0 + z3)>>1;
1892 block[stride*2 +offset]= (z1 + z2)>>1;
1893 block[stride*8 +offset]= (z1 - z2)>>1;
1894 block[stride*10+offset]= (z0 - z3)>>1;
1895 }
1896}
e5017ab8
LA
1897#endif
1898
0da71265
MN
1899#undef xStride
1900#undef stride
1901
239ea04c 1902static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1903 const int stride= 16*2;
1904 const int xStride= 16;
1905 int a,b,c,d,e;
1906
1907 a= block[stride*0 + xStride*0];
1908 b= block[stride*0 + xStride*1];
1909 c= block[stride*1 + xStride*0];
1910 d= block[stride*1 + xStride*1];
1911
1912 e= a-b;
1913 a= a+b;
1914 b= c-d;
1915 c= c+d;
1916
239ea04c
LM
1917 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1918 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1919 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1920 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
0da71265
MN
1921}
1922
e5017ab8 1923#if 0
0da71265
MN
1924static void chroma_dc_dct_c(DCTELEM *block){
1925 const int stride= 16*2;
1926 const int xStride= 16;
1927 int a,b,c,d,e;
1928
1929 a= block[stride*0 + xStride*0];
1930 b= block[stride*0 + xStride*1];
1931 c= block[stride*1 + xStride*0];
1932 d= block[stride*1 + xStride*1];
1933
1934 e= a-b;
1935 a= a+b;
1936 b= c-d;
1937 c= c+d;
1938
1939 block[stride*0 + xStride*0]= (a+c);
1940 block[stride*0 + xStride*1]= (e+b);
1941 block[stride*1 + xStride*0]= (a-c);
1942 block[stride*1 + xStride*1]= (e-b);
1943}
e5017ab8 1944#endif
0da71265
MN
1945
1946/**
1947 * gets the chroma qp.
1948 */
6ba71fc4 1949static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
115329f1 1950
6ba71fc4 1951 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
0da71265
MN
1952}
1953
3b66c4c5 1954//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
0afd2a92
DB
1955//FIXME check that gcc inlines this (and optimizes intra & separate_dc stuff away)
1956static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int separate_dc){
0da71265
MN
1957 int i;
1958 const int * const quant_table= quant_coeff[qscale];
1959 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1960 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1961 const unsigned int threshold2= (threshold1<<1);
1962 int last_non_zero;
1963
0afd2a92 1964 if(separate_dc){
0da71265
MN
1965 if(qscale<=18){
1966 //avoid overflows
1967 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1968 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1969 const unsigned int dc_threshold2= (dc_threshold1<<1);
1970
1971 int level= block[0]*quant_coeff[qscale+18][0];
1972 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1973 if(level>0){
1974 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1975 block[0]= level;
1976 }else{
1977 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1978 block[0]= -level;
1979 }
1980// last_non_zero = i;
1981 }else{
1982 block[0]=0;
1983 }
1984 }else{
1985 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1986 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1987 const unsigned int dc_threshold2= (dc_threshold1<<1);
1988
1989 int level= block[0]*quant_table[0];
1990 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1991 if(level>0){
1992 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1993 block[0]= level;
1994 }else{
1995 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1996 block[0]= -level;
1997 }
1998// last_non_zero = i;
1999 }else{
2000 block[0]=0;
2001 }
2002 }
2003 last_non_zero= 0;
2004 i=1;
2005 }else{
2006 last_non_zero= -1;
2007 i=0;
2008 }
2009
2010 for(; i<16; i++){
2011 const int j= scantable[i];
2012 int level= block[j]*quant_table[j];
2013
2014// if( bias+level >= (1<<(QMAT_SHIFT - 3))
2015// || bias-level >= (1<<(QMAT_SHIFT - 3))){
2016 if(((unsigned)(level+threshold1))>threshold2){
2017 if(level>0){
2018 level= (bias + level)>>QUANT_SHIFT;
2019 block[j]= level;
2020 }else{
2021 level= (bias - level)>>QUANT_SHIFT;
2022 block[j]= -level;
2023 }
2024 last_non_zero = i;
2025 }else{
2026 block[j]=0;
2027 }
2028 }
2029
2030 return last_non_zero;
2031}
2032
2033static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
2034 const uint32_t a= ((uint32_t*)(src-stride))[0];
2035 ((uint32_t*)(src+0*stride))[0]= a;
2036 ((uint32_t*)(src+1*stride))[0]= a;
2037 ((uint32_t*)(src+2*stride))[0]= a;
2038 ((uint32_t*)(src+3*stride))[0]= a;
2039}
2040
2041static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
2042 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
2043 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
2044 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
2045 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
2046}
2047
2048static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
2049 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
2050 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
115329f1
DB
2051
2052 ((uint32_t*)(src+0*stride))[0]=
2053 ((uint32_t*)(src+1*stride))[0]=
2054 ((uint32_t*)(src+2*stride))[0]=
2055 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
2056}
2057
2058static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
2059 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
115329f1
DB
2060
2061 ((uint32_t*)(src+0*stride))[0]=
2062 ((uint32_t*)(src+1*stride))[0]=
2063 ((uint32_t*)(src+2*stride))[0]=
2064 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
2065}
2066
2067static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
2068 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
115329f1
DB
2069
2070 ((uint32_t*)(src+0*stride))[0]=
2071 ((uint32_t*)(src+1*stride))[0]=
2072 ((uint32_t*)(src+2*stride))[0]=
2073 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
2074}
2075
2076static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2077 ((uint32_t*)(src+0*stride))[0]=
2078 ((uint32_t*)(src+1*stride))[0]=
2079 ((uint32_t*)(src+2*stride))[0]=
0da71265
MN
2080 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
2081}
2082
2083
2084#define LOAD_TOP_RIGHT_EDGE\
2085 const int t4= topright[0];\
2086 const int t5= topright[1];\
2087 const int t6= topright[2];\
2088 const int t7= topright[3];\
2089
2090#define LOAD_LEFT_EDGE\
2091 const int l0= src[-1+0*stride];\
2092 const int l1= src[-1+1*stride];\
2093 const int l2= src[-1+2*stride];\
2094 const int l3= src[-1+3*stride];\
2095
2096#define LOAD_TOP_EDGE\
2097 const int t0= src[ 0-1*stride];\
2098 const int t1= src[ 1-1*stride];\
2099 const int t2= src[ 2-1*stride];\
2100 const int t3= src[ 3-1*stride];\
2101
2102static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
2103 const int lt= src[-1-1*stride];
2104 LOAD_TOP_EDGE
2105 LOAD_LEFT_EDGE
2106
115329f1 2107 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
0da71265 2108 src[0+2*stride]=
115329f1 2109 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
0da71265
MN
2110 src[0+1*stride]=
2111 src[1+2*stride]=
115329f1 2112 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
0da71265
MN
2113 src[0+0*stride]=
2114 src[1+1*stride]=
2115 src[2+2*stride]=
115329f1 2116 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
0da71265
MN
2117 src[1+0*stride]=
2118 src[2+1*stride]=
2119 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
2120 src[2+0*stride]=
2121 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2122 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
4cfbf61b 2123}
0da71265
MN
2124
2125static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2126 LOAD_TOP_EDGE
2127 LOAD_TOP_RIGHT_EDGE
2128// LOAD_LEFT_EDGE
0da71265
MN
2129
2130 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
2131 src[1+0*stride]=
2132 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
2133 src[2+0*stride]=
2134 src[1+1*stride]=
2135 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
2136 src[3+0*stride]=
2137 src[2+1*stride]=
2138 src[1+2*stride]=
2139 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
2140 src[3+1*stride]=
2141 src[2+2*stride]=
2142 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
2143 src[3+2*stride]=
2144 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
2145 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
4cfbf61b 2146}
0da71265
MN
2147
2148static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
2149 const int lt= src[-1-1*stride];
115329f1
DB
2150 LOAD_TOP_EDGE
2151 LOAD_LEFT_EDGE
0da71265
MN
2152 const __attribute__((unused)) int unu= l3;
2153
2154 src[0+0*stride]=
2155 src[1+2*stride]=(lt + t0 + 1)>>1;
2156 src[1+0*stride]=
2157 src[2+2*stride]=(t0 + t1 + 1)>>1;
2158 src[2+0*stride]=
2159 src[3+2*stride]=(t1 + t2 + 1)>>1;
2160 src[3+0*stride]=(t2 + t3 + 1)>>1;
2161 src[0+1*stride]=
2162 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2163 src[1+1*stride]=
2164 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2165 src[2+1*stride]=
2166 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2167 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2168 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2169 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
4cfbf61b 2170}
0da71265
MN
2171
2172static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2173 LOAD_TOP_EDGE
2174 LOAD_TOP_RIGHT_EDGE
0da71265
MN
2175 const __attribute__((unused)) int unu= t7;
2176
2177 src[0+0*stride]=(t0 + t1 + 1)>>1;
2178 src[1+0*stride]=
2179 src[0+2*stride]=(t1 + t2 + 1)>>1;
2180 src[2+0*stride]=
2181 src[1+2*stride]=(t2 + t3 + 1)>>1;
2182 src[3+0*stride]=
2183 src[2+2*stride]=(t3 + t4+ 1)>>1;
2184 src[3+2*stride]=(t4 + t5+ 1)>>1;
2185 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2186 src[1+1*stride]=
2187 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2188 src[2+1*stride]=
2189 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2190 src[3+1*stride]=
2191 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2192 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
4cfbf61b 2193}
0da71265
MN
2194
2195static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
115329f1 2196 LOAD_LEFT_EDGE
0da71265
MN
2197
2198 src[0+0*stride]=(l0 + l1 + 1)>>1;
2199 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2200 src[2+0*stride]=
2201 src[0+1*stride]=(l1 + l2 + 1)>>1;
2202 src[3+0*stride]=
2203 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2204 src[2+1*stride]=
2205 src[0+2*stride]=(l2 + l3 + 1)>>1;
2206 src[3+1*stride]=
2207 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2208 src[3+2*stride]=
2209 src[1+3*stride]=
2210 src[0+3*stride]=
2211 src[2+2*stride]=
2212 src[2+3*stride]=
2213 src[3+3*stride]=l3;
4cfbf61b 2214}
115329f1 2215
0da71265
MN
2216static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2217 const int lt= src[-1-1*stride];
115329f1
DB
2218 LOAD_TOP_EDGE
2219 LOAD_LEFT_EDGE
0da71265
MN
2220 const __attribute__((unused)) int unu= t3;
2221
2222 src[0+0*stride]=
2223 src[2+1*stride]=(lt + l0 + 1)>>1;
2224 src[1+0*stride]=
2225 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2226 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2227 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2228 src[0+1*stride]=
2229 src[2+2*stride]=(l0 + l1 + 1)>>1;
2230 src[1+1*stride]=
2231 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2232 src[0+2*stride]=
2233 src[2+3*stride]=(l1 + l2+ 1)>>1;
2234 src[1+2*stride]=
2235 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2236 src[0+3*stride]=(l2 + l3 + 1)>>1;
2237 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
4cfbf61b 2238}
0da71265 2239
a074f143 2240void ff_pred16x16_vertical_c(uint8_t *src, int stride){
0da71265
MN
2241 int i;
2242 const uint32_t a= ((uint32_t*)(src-stride))[0];
2243 const uint32_t b= ((uint32_t*)(src-stride))[1];
2244 const uint32_t c= ((uint32_t*)(src-stride))[2];
2245 const uint32_t d= ((uint32_t*)(src-stride))[3];
115329f1 2246
0da71265
MN
2247 for(i=0; i<16; i++){
2248 ((uint32_t*)(src+i*stride))[0]= a;
2249 ((uint32_t*)(src+i*stride))[1]= b;
2250 ((uint32_t*)(src+i*stride))[2]= c;
2251 ((uint32_t*)(src+i*stride))[3]= d;
2252 }
2253}
2254
a074f143 2255void ff_pred16x16_horizontal_c(uint8_t *src, int stride){
0da71265
MN
2256 int i;
2257
2258 for(i=0; i<16; i++){
2259 ((uint32_t*)(src+i*stride))[0]=
2260 ((uint32_t*)(src+i*stride))[1]=
2261 ((uint32_t*)(src+i*stride))[2]=
2262 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2263 }
2264}
2265
a074f143 2266void ff_pred16x16_dc_c(uint8_t *src, int stride){
0da71265
MN
2267 int i, dc=0;
2268
2269 for(i=0;i<16; i++){
2270 dc+= src[-1+i*stride];
2271 }
115329f1 2272
0da71265
MN
2273 for(i=0;i<16; i++){
2274 dc+= src[i-stride];
2275 }
2276
2277 dc= 0x01010101*((dc + 16)>>5);
2278
2279 for(i=0; i<16; i++){
2280 ((uint32_t*)(src+i*stride))[0]=
2281 ((uint32_t*)(src+i*stride))[1]=
2282 ((uint32_t*)(src+i*stride))[2]=
2283 ((uint32_t*)(src+i*stride))[3]= dc;
2284 }
2285}
2286
2287static void pred16x16_left_dc_c(uint8_t *src, int stride){
2288 int i, dc=0;
2289
2290 for(i=0;i<16; i++){
2291 dc+= src[-1+i*stride];
2292 }
115329f1 2293
0da71265
MN
2294 dc= 0x01010101*((dc + 8)>>4);
2295
2296 for(i=0; i<16; i++){
2297 ((uint32_t*)(src+i*stride))[0]=
2298 ((uint32_t*)(src+i*stride))[1]=
2299 ((uint32_t*)(src+i*stride))[2]=
2300 ((uint32_t*)(src+i*stride))[3]= dc;
2301 }
2302}
2303
2304static void pred16x16_top_dc_c(uint8_t *src, int stride){
2305 int i, dc=0;
2306
2307 for(i=0;i<16; i++){
2308 dc+= src[i-stride];
2309 }
2310 dc= 0x01010101*((dc + 8)>>4);
2311
2312 for(i=0; i<16; i++){
2313 ((uint32_t*)(src+i*stride))[0]=
2314 ((uint32_t*)(src+i*stride))[1]=
2315 ((uint32_t*)(src+i*stride))[2]=
2316 ((uint32_t*)(src+i*stride))[3]= dc;
2317 }
2318}
2319
a074f143 2320void ff_pred16x16_128_dc_c(uint8_t *src, int stride){
0da71265
MN
2321 int i;
2322
2323 for(i=0; i<16; i++){
2324 ((uint32_t*)(src+i*stride))[0]=
2325 ((uint32_t*)(src+i*stride))[1]=
2326 ((uint32_t*)(src+i*stride))[2]=
2327 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2328 }
2329}
2330
8b82a956 2331static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
30f73fc7
MN
2332 int i, j, k;
2333 int a;
55fde95e 2334 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
30f73fc7
MN
2335 const uint8_t * const src0 = src+7-stride;
2336 const uint8_t *src1 = src+8*stride-1;
2337 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2338 int H = src0[1] - src0[-1];
2339 int V = src1[0] - src2[ 0];
2340 for(k=2; k<=8; ++k) {
2341 src1 += stride; src2 -= stride;
2342 H += k*(src0[k] - src0[-k]);
2343 V += k*(src1[0] - src2[ 0]);
2344 }
8b82a956
MN
2345 if(svq3){
2346 H = ( 5*(H/4) ) / 16;
2347 V = ( 5*(V/4) ) / 16;
2e26c8d2
MM
2348
2349 /* required for 100% accuracy */
2350 i = H; H = V; V = i;
8b82a956
MN
2351 }else{
2352 H = ( 5*H+32 ) >> 6;
2353 V = ( 5*V+32 ) >> 6;
2354 }
30f73fc7
MN
2355
2356 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2357 for(j=16; j>0; --j) {
2358 int b = a;
2359 a += V;
2360 for(i=-16; i<0; i+=4) {
2361 src[16+i] = cm[ (b ) >> 5 ];
2362 src[17+i] = cm[ (b+ H) >> 5 ];
2363 src[18+i] = cm[ (b+2*H) >> 5 ];
2364 src[19+i] = cm[ (b+3*H) >> 5 ];
2365 b += 4*H;
0da71265 2366 }
30f73fc7
MN
2367 src += stride;
2368 }
0da71265
MN
2369}
2370
a074f143 2371void ff_pred16x16_plane_c(uint8_t *src, int stride){
8b82a956
MN
2372 pred16x16_plane_compat_c(src, stride, 0);
2373}
2374
a074f143 2375void ff_pred8x8_vertical_c(uint8_t *src, int stride){
0da71265
MN
2376 int i;
2377 const uint32_t a= ((uint32_t*)(src-stride))[0];
2378 const uint32_t b= ((uint32_t*)(src-stride))[1];
115329f1 2379
0da71265
MN
2380 for(i=0; i<8; i++){
2381 ((uint32_t*)(src+i*stride))[0]= a;
2382 ((uint32_t*)(src+i*stride))[1]= b;
2383 }
2384}
2385
a074f143 2386void ff_pred8x8_horizontal_c(uint8_t *src, int stride){
0da71265
MN
2387 int i;
2388
2389 for(i=0; i<8; i++){
2390 ((uint32_t*)(src+i*stride))[0]=
2391 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2392 }
2393}
2394
a074f143 2395void ff_pred8x8_128_dc_c(uint8_t *src, int stride){
0da71265
MN
2396 int i;
2397
43efd19a 2398 for(i=0; i<8; i++){
115329f1 2399 ((uint32_t*)(src+i*stride))[0]=
0da71265
MN
2400 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2401 }
2402}
2403
2404static void pred8x8_left_dc_c(uint8_t *src, int stride){
2405 int i;
2406 int dc0, dc2;
2407
2408 dc0=dc2=0;
2409 for(i=0;i<4; i++){
2410 dc0+= src[-1+i*stride];
2411 dc2+= src[-1+(i+4)*stride];
2412 }
2413 dc0= 0x01010101*((dc0 + 2)>>2);
2414 dc2= 0x01010101*((dc2 + 2)>>2);
2415
2416 for(i=0; i<4; i++){
2417 ((uint32_t*)(src+i*stride))[0]=
2418 ((uint32_t*)(src+i*stride))[1]= dc0;
2419 }
2420 for(i=4; i<8; i++){
2421 ((uint32_t*)(src+i*stride))[0]=
2422 ((uint32_t*)(src+i*stride))[1]= dc2;
2423 }
2424}
2425
2426static void pred8x8_top_dc_c(uint8_t *src, int stride){
2427 int i;
2428 int dc0, dc1;
2429
2430 dc0=dc1=0;
2431 for(i=0;i<4; i++){
2432 dc0+= src[i-stride];
2433 dc1+= src[4+i-stride];
2434 }
2435 dc0= 0x01010101*((dc0 + 2)>>2);
2436 dc1= 0x01010101*((dc1 + 2)>>2);
2437
2438 for(i=0; i<4; i++){
2439 ((uint32_t*)(src+i*stride))[0]= dc0;
2440 ((uint32_t*)(src+i*stride))[1]= dc1;
2441 }
2442 for(i=4; i<8; i++){
2443 ((uint32_t*)(src+i*stride))[0]= dc0;
2444 ((uint32_t*)(src+i*stride))[1]= dc1;
2445 }
2446}
2447
2448
a074f143 2449void ff_pred8x8_dc_c(uint8_t *src, int stride){
0da71265
MN
2450 int i;
2451 int dc0, dc1, dc2, dc3;
2452
2453 dc0=dc1=dc2=0;
2454 for(i=0;i<4; i++){
2455 dc0+= src[-1+i*stride] + src[i-stride];
2456 dc1+= src[4+i-stride];
2457 dc2+= src[-1+(i+4)*stride];
2458 }
2459 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2460 dc0= 0x01010101*((dc0 + 4)>>3);
2461 dc1= 0x01010101*((dc1 + 2)>>2);
2462 dc2= 0x01010101*((dc2 + 2)>>2);
2463
2464 for(i=0; i<4; i++){
2465 ((uint32_t*)(src+i*stride))[0]= dc0;
2466 ((uint32_t*)(src+i*stride))[1]= dc1;
2467 }
2468 for(i=4; i<8; i++){
2469 ((uint32_t*)(src+i*stride))[0]= dc2;
2470 ((uint32_t*)(src+i*stride))[1]= dc3;
2471 }
2472}
2473
a074f143 2474void ff_pred8x8_plane_c(uint8_t *src, int stride){
30f73fc7
MN
2475 int j, k;
2476 int a;
55fde95e 2477 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
30f73fc7
MN
2478 const uint8_t * const src0 = src+3-stride;
2479 const uint8_t *src1 = src+4*stride-1;
2480 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2481 int H = src0[1] - src0[-1];
2482 int V = src1[0] - src2[ 0];
2483 for(k=2; k<=4; ++k) {
2484 src1 += stride; src2 -= stride;
2485 H += k*(src0[k] - src0[-k]);
2486 V += k*(src1[0] - src2[ 0]);
2487 }
2488 H = ( 17*H+16 ) >> 5;
2489 V = ( 17*V+16 ) >> 5;
2490
2491 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2492 for(j=8; j>0; --j) {
2493 int b = a;
2494 a += V;
2495 src[0] = cm[ (b ) >> 5 ];
2496 src[1] = cm[ (b+ H) >> 5 ];
2497 src[2] = cm[ (b+2*H) >> 5 ];
2498 src[3] = cm[ (b+3*H) >> 5 ];
2499 src[4] = cm[ (b+4*H) >> 5 ];
2500 src[5] = cm[ (b+5*H) >> 5 ];
2501 src[6] = cm[ (b+6*H) >> 5 ];
2502 src[7] = cm[ (b+7*H) >> 5 ];
2503 src += stride;
2504 }
0da71265
MN
2505}
2506
43efd19a
LM
2507#define SRC(x,y) src[(x)+(y)*stride]
2508#define PL(y) \
2509 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2510#define PREDICT_8x8_LOAD_LEFT \
2511 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2512 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2513 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
bf4e3bd2 2514 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
43efd19a
LM
2515
2516#define PT(x) \
2517 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2518#define PREDICT_8x8_LOAD_TOP \
2519 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2520 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2521 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
bf4e3bd2 2522 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
fb3ca6ba 2523 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
43efd19a
LM
2524
2525#define PTR(x) \
2526 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2527#define PREDICT_8x8_LOAD_TOPRIGHT \
2528 int t8, t9, t10, t11, t12, t13, t14, t15; \
2529 if(has_topright) { \
2530 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2531 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2532 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2533
2534#define PREDICT_8x8_LOAD_TOPLEFT \
fb3ca6ba 2535 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
43efd19a
LM
2536
2537#define PREDICT_8x8_DC(v) \
2538 int y; \
2539 for( y = 0; y < 8; y++ ) { \
2540 ((uint32_t*)src)[0] = \
2541 ((uint32_t*)src)[1] = v; \
2542 src += stride; \
2543 }
2544
2545static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2546{
2547 PREDICT_8x8_DC(0x80808080);
2548}
2549static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2550{
2551 PREDICT_8x8_LOAD_LEFT;
2552 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2553 PREDICT_8x8_DC(dc);
2554}
2555static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2556{
2557 PREDICT_8x8_LOAD_TOP;
2558 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2559 PREDICT_8x8_DC(dc);
2560}
2561static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2562{
2563 PREDICT_8x8_LOAD_LEFT;
2564 PREDICT_8x8_LOAD_TOP;
2565 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2566 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2567 PREDICT_8x8_DC(dc);
2568}
2569static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2570{
2571 PREDICT_8x8_LOAD_LEFT;
2572#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2573 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2574 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2575#undef ROW
2576}
2577static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2578{
2579 int y;
2580 PREDICT_8x8_LOAD_TOP;
2581 src[0] = t0;
2582 src[1] = t1;
2583 src[2] = t2;
2584 src[3] = t3;
2585 src[4] = t4;
2586 src[5] = t5;
2587 src[6] = t6;
2588 src[7] = t7;
2589 for( y = 1; y < 8; y++ )
2590 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2591}
2592static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2593{
2594 PREDICT_8x8_LOAD_TOP;
2595 PREDICT_8x8_LOAD_TOPRIGHT;
2596 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2597 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2598 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2599 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2600 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2601 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2602 SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
2603 SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
2604 SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
2605 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2606 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2607 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2608 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2609 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2610 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2611}
2612static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2613{
2614 PREDICT_8x8_LOAD_TOP;
2615 PREDICT_8x8_LOAD_LEFT;
2616 PREDICT_8x8_LOAD_TOPLEFT;
2617 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2618 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2619 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2620 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2621 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2622 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2623 SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
2624 SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
2625 SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
2626 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2627 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2628 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2629 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2630 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2631 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
115329f1 2632
43efd19a
LM
2633}
2634static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2635{
2636 PREDICT_8x8_LOAD_TOP;
2637 PREDICT_8x8_LOAD_LEFT;
2638 PREDICT_8x8_LOAD_TOPLEFT;
2639 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2640 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2641 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2642 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2643 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2644 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2645 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2646 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2647 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2648 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2649 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2650 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2651 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2652 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2653 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2654 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2655 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2656 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2657 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2658 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2659 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2660 SRC(7,0)= (t6 + t7 + 1) >> 1;
2661}
2662static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2663{
2664 PREDICT_8x8_LOAD_TOP;
2665 PREDICT_8x8_LOAD_LEFT;
2666 PREDICT_8x8_LOAD_TOPLEFT;
2667 SRC(0,7)= (l6 + l7 + 1) >> 1;
2668 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2669 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2670 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2671 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2672 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2673 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2674 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2675 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2676 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2677 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2678 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2679 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2680 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2681 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2682 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2683 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2684 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2685 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2686 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2687 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2688 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2689}
2690static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2691{
2692 PREDICT_8x8_LOAD_TOP;
2693 PREDICT_8x8_LOAD_TOPRIGHT;
2694 SRC(0,0)= (t0 + t1 + 1) >> 1;
2695 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2696 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2697 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2698 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2699 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2700 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2701 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2702 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2703 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2704 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2705 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2706 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2707 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2708 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2709 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2710 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2711 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2712 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2713 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2714 SRC(7,6)= (t10 + t11 + 1) >> 1;
2715 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2716}
2717static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2718{
2719 PREDICT_8x8_LOAD_LEFT;
2720 SRC(0,0)= (l0 + l1 + 1) >> 1;
2721 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2722 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2723 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2724 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2725 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2726 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2727 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2728 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2729 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2730 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2731 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2732 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2733 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2734 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2735 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2736 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2737 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2738}
2739#undef PREDICT_8x8_LOAD_LEFT
2740#undef PREDICT_8x8_LOAD_TOP
2741#undef PREDICT_8x8_LOAD_TOPLEFT
2742#undef PREDICT_8x8_LOAD_TOPRIGHT
2743#undef PREDICT_8x8_DC
2744#undef PTR
2745#undef PT
2746#undef PL
2747#undef SRC
2748
0da71265
MN
2749static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2750 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2751 int src_x_offset, int src_y_offset,
2752 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2753 MpegEncContext * const s = &h->s;
2754 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
5d18eaad 2755 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
0da71265 2756 const int luma_xy= (mx&3) + ((my&3)<<2);
5d18eaad
LM
2757 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*h->mb_linesize;
2758 uint8_t * src_cb, * src_cr;
2759 int extra_width= h->emu_edge_width;
2760 int extra_height= h->emu_edge_height;
0da71265
MN
2761 int emu=0;
2762 const int full_mx= mx>>2;
2763 const int full_my= my>>2;
fbd312fd 2764 const int pic_width = 16*s->mb_width;
5d18eaad 2765 const int pic_height = 16*s->mb_height >> MB_MBAFF;
115329f1 2766
2f29af39 2767 if(!pic->data[0]) //FIXME this is unacceptable, some senseable error concealment must be done for missing reference frames
171c4076 2768 return;
115329f1 2769
0da71265
MN
2770 if(mx&7) extra_width -= 3;
2771 if(my&7) extra_height -= 3;
115329f1
DB
2772
2773 if( full_mx < 0-extra_width
2774 || full_my < 0-extra_height
2775 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd 2776 || full_my + 16/*FIXME*/ > pic_height + extra_height){
5d18eaad
LM
2777 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
2778 src_y= s->edge_emu_buffer + 2 + 2*h->mb_linesize;
0da71265
MN
2779 emu=1;
2780 }
115329f1 2781
5d18eaad 2782 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
0da71265 2783 if(!square){
5d18eaad 2784 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
0da71265 2785 }
115329f1 2786
0da71265 2787 if(s->flags&CODEC_FLAG_GRAY) return;
115329f1 2788
5d18eaad
LM
2789 if(MB_MBAFF){
2790 // chroma offset when predicting from a field of opposite parity
2791 my += 2 * ((s->mb_y & 1) - (h->ref_cache[list][scan8[n]] & 1));
2792 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
2793 }
2794 src_cb= pic->data[1] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2795 src_cr= pic->data[2] + (mx>>3) + (my>>3)*h->mb_uvlinesize;
2796
0da71265 2797 if(emu){
5d18eaad 2798 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
2799 src_cb= s->edge_emu_buffer;
2800 }
5d18eaad 2801 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
2802
2803 if(emu){
5d18eaad 2804 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
2805 src_cr= s->edge_emu_buffer;
2806 }
5d18eaad 2807 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
0da71265
MN
2808}
2809
9f2d1b4f 2810static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
2811 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2812 int x_offset, int y_offset,
2813 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2814 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2815 int list0, int list1){
2816 MpegEncContext * const s = &h->s;
2817 qpel_mc_func *qpix_op= qpix_put;
2818 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 2819
5d18eaad
LM
2820 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2821 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2822 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
0da71265 2823 x_offset += 8*s->mb_x;
5d18eaad 2824 y_offset += 8*(s->mb_y >> MB_MBAFF);
115329f1 2825
0da71265 2826 if(list0){
1924f3ce 2827 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
2828 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2829 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2830 qpix_op, chroma_op);
2831
2832 qpix_op= qpix_avg;
2833 chroma_op= chroma_avg;
2834 }
2835
2836 if(list1){
1924f3ce 2837 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
2838 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2839 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2840 qpix_op, chroma_op);
2841 }
2842}
2843
9f2d1b4f
LM
2844static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2845 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2846 int x_offset, int y_offset,
2847 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2848 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2849 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2850 int list0, int list1){
2851 MpegEncContext * const s = &h->s;
2852
5d18eaad
LM
2853 dest_y += 2*x_offset + 2*y_offset*h-> mb_linesize;
2854 dest_cb += x_offset + y_offset*h->mb_uvlinesize;
2855 dest_cr += x_offset + y_offset*h->mb_uvlinesize;
9f2d1b4f 2856 x_offset += 8*s->mb_x;
5d18eaad 2857 y_offset += 8*(s->mb_y >> MB_MBAFF);
115329f1 2858
9f2d1b4f
LM
2859 if(list0 && list1){
2860 /* don't optimize for luma-only case, since B-frames usually
2861 * use implicit weights => chroma too. */
2862 uint8_t *tmp_cb = s->obmc_scratchpad;
5d18eaad
LM
2863 uint8_t *tmp_cr = s->obmc_scratchpad + 8;
2864 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
9f2d1b4f
LM
2865 int refn0 = h->ref_cache[0][ scan8[n] ];
2866 int refn1 = h->ref_cache[1][ scan8[n] ];
2867
2868 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2869 dest_y, dest_cb, dest_cr,
2870 x_offset, y_offset, qpix_put, chroma_put);
2871 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2872 tmp_y, tmp_cb, tmp_cr,
2873 x_offset, y_offset, qpix_put, chroma_put);
2874
2875 if(h->use_weight == 2){
2876 int weight0 = h->implicit_weight[refn0][refn1];
2877 int weight1 = 64 - weight0;
5d18eaad
LM
2878 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
2879 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
2880 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f 2881 }else{
5d18eaad 2882 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
115329f1 2883 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
e8b56208 2884 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
5d18eaad 2885 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
115329f1 2886 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
e8b56208 2887 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
5d18eaad 2888 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
115329f1 2889 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
e8b56208 2890 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
9f2d1b4f
LM
2891 }
2892 }else{
2893 int list = list1 ? 1 : 0;
2894 int refn = h->ref_cache[list][ scan8[n] ];
2895 Picture *ref= &h->ref_list[list][refn];
2896 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2897 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2898 qpix_put, chroma_put);
2899
5d18eaad 2900 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
9f2d1b4f
LM
2901 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2902 if(h->use_weight_chroma){
5d18eaad 2903 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
9f2d1b4f 2904 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
5d18eaad 2905 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
9f2d1b4f
LM
2906 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2907 }
2908 }
2909}
2910
2911static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2912 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2913 int x_offset, int y_offset,
2914 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2915 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 2916 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
9f2d1b4f
LM
2917 int list0, int list1){
2918 if((h->use_weight==2 && list0 && list1
2919 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2920 || h->use_weight==1)
2921 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2922 x_offset, y_offset, qpix_put, chroma_put,
2923 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2924 else
2925 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2926 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2927}
2928
513fbd8e
LM
2929static inline void prefetch_motion(H264Context *h, int list){
2930 /* fetch pixels for estimated mv 4 macroblocks ahead
2931 * optimized for 64byte cache lines */
2932 MpegEncContext * const s = &h->s;
2933 const int refn = h->ref_cache[list][scan8[0]];
2934 if(refn >= 0){
2935 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2936 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2937 uint8_t **src= h->ref_list[list][refn].data;
5d18eaad 2938 int off= mx + (my + (s->mb_x&3)*4)*h->mb_linesize + 64;
513fbd8e
LM
2939 s->dsp.prefetch(src[0]+off, s->linesize, 4);
2940 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2941 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2942 }
2943}
2944
0da71265
MN
2945static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2946 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
2947 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2948 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 2949 MpegEncContext * const s = &h->s;
7bc9090a 2950 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 2951 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 2952
0da71265 2953 assert(IS_INTER(mb_type));
115329f1 2954
513fbd8e
LM
2955 prefetch_motion(h, 0);
2956
0da71265
MN
2957 if(IS_16X16(mb_type)){
2958 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2959 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 2960 &weight_op[0], &weight_avg[0],
0da71265
MN
2961 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2962 }else if(IS_16X8(mb_type)){
2963 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2964 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2965 &weight_op[1], &weight_avg[1],
0da71265
MN
2966 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2967 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2968 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2969 &weight_op[1], &weight_avg[1],
0da71265
MN
2970 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2971 }else if(IS_8X16(mb_type)){
5d18eaad 2972 mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
0da71265 2973 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2974 &weight_op[2], &weight_avg[2],
0da71265 2975 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
5d18eaad 2976 mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
0da71265 2977 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2978 &weight_op[2], &weight_avg[2],
0da71265
MN
2979 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2980 }else{
2981 int i;
115329f1 2982
0da71265
MN
2983 assert(IS_8X8(mb_type));
2984
2985 for(i=0; i<4; i++){
2986 const int sub_mb_type= h->sub_mb_type[i];
2987 const int n= 4*i;
2988 int x_offset= (i&1)<<2;
2989 int y_offset= (i&2)<<1;
2990
2991 if(IS_SUB_8X8(sub_mb_type)){
2992 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2993 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2994 &weight_op[3], &weight_avg[3],
0da71265
MN
2995 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2996 }else if(IS_SUB_8X4(sub_mb_type)){
2997 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2998 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2999 &weight_op[4], &weight_avg[4],
0da71265
MN
3000 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3001 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
3002 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 3003 &weight_op[4], &weight_avg[4],
0da71265
MN
3004 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3005 }else if(IS_SUB_4X8(sub_mb_type)){
5d18eaad 3006 mc_part(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
0da71265 3007 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 3008 &weight_op[5], &weight_avg[5],
0da71265 3009 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
5d18eaad 3010 mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
0da71265 3011 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 3012 &weight_op[5], &weight_avg[5],
0da71265
MN
3013 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3014 }else{
3015 int j;
3016 assert(IS_SUB_4X4(sub_mb_type));
3017 for(j=0; j<4; j++){
3018 int sub_x_offset= x_offset + 2*(j&1);
3019 int sub_y_offset= y_offset + (j&2);
3020 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
3021 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 3022 &weight_op[6], &weight_avg[6],
0da71265
MN
3023 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
3024 }
3025 }
3026 }
3027 }
513fbd8e
LM
3028
3029 prefetch_motion(h, 1);
0da71265
MN
3030}
3031
c2212338 3032static void decode_init_vlc(){
0da71265
MN
3033 static int done = 0;
3034
3035 if (!done) {
3036 int i;
3037 done = 1;
3038
115329f1 3039 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
0da71265 3040 &chroma_dc_coeff_token_len [0], 1, 1,
073c2593 3041 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
0da71265
MN
3042
3043 for(i=0; i<4; i++){
115329f1 3044 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
0da71265 3045 &coeff_token_len [i][0], 1, 1,
073c2593 3046 &coeff_token_bits[i][0], 1, 1, 1);
0da71265
MN
3047 }
3048
3049 for(i=0; i<3; i++){
3050 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
3051 &chroma_dc_total_zeros_len [i][0], 1, 1,
073c2593 3052 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
3053 }
3054 for(i=0; i<15; i++){
115329f1 3055 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
0da71265 3056 &total_zeros_len [i][0], 1, 1,
073c2593 3057 &total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
3058 }
3059
3060 for(i=0; i<6; i++){
115329f1 3061 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
0da71265 3062 &run_len [i][0], 1, 1,
073c2593 3063 &run_bits[i][0], 1, 1, 1);
0da71265 3064 }
115329f1 3065 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
0da71265 3066 &run_len [6][0], 1, 1,
073c2593 3067 &run_bits[6][0], 1, 1, 1);
0da71265
MN
3068 }
3069}
3070
3071/**
3072 * Sets the intra prediction function pointers.
3073 */
3074static void init_pred_ptrs(H264Context *h){
3075// MpegEncContext * const s = &h->s;
3076
3077 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
3078 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
3079 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
3080 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
3081 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
3082 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
3083 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
3084 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
3085 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
3086 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
3087 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
3088 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
3089
43efd19a
LM
3090 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
3091 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
3092 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
3093 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
3094 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
3095 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
3096 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
3097 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
3098 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
3099 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
3100 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
3101 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
3102
a074f143
PI
3103 h->pred8x8[DC_PRED8x8 ]= ff_pred8x8_dc_c;
3104 h->pred8x8[VERT_PRED8x8 ]= ff_pred8x8_vertical_c;
3105 h->pred8x8[HOR_PRED8x8 ]= ff_pred8x8_horizontal_c;
3106 h->pred8x8[PLANE_PRED8x8 ]= ff_pred8x8_plane_c;
0da71265
MN
3107 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
3108 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
a074f143 3109 h->pred8x8[DC_128_PRED8x8 ]= ff_pred8x8_128_dc_c;
0da71265 3110
a074f143
PI
3111 h->pred16x16[DC_PRED8x8 ]= ff_pred16x16_dc_c;
3112 h->pred16x16[VERT_PRED8x8 ]= ff_pred16x16_vertical_c;
3113 h->pred16x16[HOR_PRED8x8 ]= ff_pred16x16_horizontal_c;
3114 h->pred16x16[PLANE_PRED8x8 ]= ff_pred16x16_plane_c;
0da71265
MN
3115 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
3116 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
a074f143 3117 h->pred16x16[DC_128_PRED8x8 ]= ff_pred16x16_128_dc_c;
0da71265
MN
3118}
3119
0da71265 3120static void free_tables(H264Context *h){
0da71265 3121 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
3122 av_freep(&h->chroma_pred_mode_table);
3123 av_freep(&h->cbp_table);
9e528114
LA
3124 av_freep(&h->mvd_table[0]);
3125 av_freep(&h->mvd_table[1]);
5ad984c9 3126 av_freep(&h->direct_table);
0da71265
MN
3127 av_freep(&h->non_zero_count);
3128 av_freep(&h->slice_table_base);
6ba71fc4
LLL
3129 av_freep(&h->top_borders[1]);
3130 av_freep(&h->top_borders[0]);
0da71265 3131 h->slice_table= NULL;
e5017ab8 3132
0da71265
MN
3133 av_freep(&h->mb2b_xy);
3134 av_freep(&h->mb2b8_xy);
9f2d1b4f
LM
3135
3136 av_freep(&h->s.obmc_scratchpad);
0da71265
MN
3137}
3138
239ea04c
LM
3139static void init_dequant8_coeff_table(H264Context *h){
3140 int i,q,x;
548a1c8a 3141 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
239ea04c
LM
3142 h->dequant8_coeff[0] = h->dequant8_buffer[0];
3143 h->dequant8_coeff[1] = h->dequant8_buffer[1];
3144
3145 for(i=0; i<2; i++ ){
3146 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
3147 h->dequant8_coeff[1] = h->dequant8_buffer[0];
3148 break;
3149 }
3150
3151 for(q=0; q<52; q++){
acd8d10f
PI
3152 int shift = ff_div6[q];
3153 int idx = ff_rem6[q];
239ea04c 3154 for(x=0; x<64; x++)
548a1c8a
LM
3155 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
3156 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
3157 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
3158 }
3159 }
3160}
3161
3162static void init_dequant4_coeff_table(H264Context *h){
3163 int i,j,q,x;
ab2e3e2c 3164 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
239ea04c
LM
3165 for(i=0; i<6; i++ ){
3166 h->dequant4_coeff[i] = h->dequant4_buffer[i];
3167 for(j=0; j<i; j++){
3168 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
3169 h->dequant4_coeff[i] = h->dequant4_buffer[j];
3170 break;
3171 }
3172 }
3173 if(j<i)
3174 continue;
3175
3176 for(q=0; q<52; q++){
acd8d10f
PI
3177 int shift = ff_div6[q] + 2;
3178 int idx = ff_rem6[q];
239ea04c 3179 for(x=0; x<16; x++)
ab2e3e2c
LM
3180 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3181 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
3182 h->pps.scaling_matrix4[i][x]) << shift;
3183 }
3184 }
3185}
3186
3187static void init_dequant_tables(H264Context *h){
3188 int i,x;
3189 init_dequant4_coeff_table(h);
3190 if(h->pps.transform_8x8_mode)
3191 init_dequant8_coeff_table(h);
3192 if(h->sps.transform_bypass){
3193 for(i=0; i<6; i++)
3194 for(x=0; x<16; x++)
3195 h->dequant4_coeff[i][0][x] = 1<<6;
3196 if(h->pps.transform_8x8_mode)
3197 for(i=0; i<2; i++)
3198 for(x=0; x<64; x++)
3199 h->dequant8_coeff[i][0][x] = 1<<6;
3200 }
3201}
3202
3203
0da71265
MN
3204/**
3205 * allocates tables.
3b66c4c5 3206 * needs width/height
0da71265
MN
3207 */
3208static int alloc_tables(H264Context *h){
3209 MpegEncContext * const s = &h->s;
7bc9090a 3210 const int big_mb_num= s->mb_stride * (s->mb_height+1);
239ea04c 3211 int x,y;
0da71265
MN
3212
3213 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 3214
53c05b1e 3215 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
5d18eaad 3216 CHECKED_ALLOCZ(h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(uint8_t))
6ba71fc4
LLL
3217 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
3218 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
5d0e4cb8 3219 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 3220
e5017ab8
LA
3221 if( h->pps.cabac ) {
3222 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
9e528114
LA
3223 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3224 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
5ad984c9 3225 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8
LA
3226 }
3227
5d18eaad
LM
3228 memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(uint8_t));
3229 h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
0da71265 3230
a55f20bd
LM
3231 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
3232 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
0da71265
MN
3233 for(y=0; y<s->mb_height; y++){
3234 for(x=0; x<s->mb_width; x++){
7bc9090a 3235 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
3236 const int b_xy = 4*x + 4*y*h->b_stride;
3237 const int b8_xy= 2*x + 2*y*h->b8_stride;
115329f1 3238
0da71265
MN
3239 h->mb2b_xy [mb_xy]= b_xy;
3240 h->mb2b8_xy[mb_xy]= b8_xy;
3241 }
3242 }
9f2d1b4f 3243
9c6221ae
GV
3244 s->obmc_scratchpad = NULL;
3245
56edbd81
LM
3246 if(!h->dequant4_coeff[0])
3247 init_dequant_tables(h);
3248
0da71265
MN
3249 return 0;
3250fail:
3251 free_tables(h);
3252 return -1;
3253}
3254
3255static void common_init(H264Context *h){
3256 MpegEncContext * const s = &h->s;
0da71265
MN
3257
3258 s->width = s->avctx->width;
3259 s->height = s->avctx->height;
3260 s->codec_id= s->avctx->codec->id;
115329f1 3261
0da71265
MN
3262 init_pred_ptrs(h);
3263
239ea04c 3264 h->dequant_coeff_pps= -1;
9a41c2c7 3265 s->unrestricted_mv=1;
0da71265 3266 s->decode=1; //FIXME
56edbd81
LM
3267
3268 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3269 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
3270}
3271
3272static int decode_init(AVCodecContext *avctx){
3273 H264Context *h= avctx->priv_data;
3274 MpegEncContext * const s = &h->s;
3275
3edcacde 3276 MPV_decode_defaults(s);
115329f1 3277
0da71265
MN
3278 s->avctx = avctx;
3279 common_init(h);
3280
3281 s->out_format = FMT_H264;
3282 s->workaround_bugs= avctx->workaround_bugs;
3283
3284 // set defaults
0da71265
MN
3285// s->decode_mb= ff_h263_decode_mb;
3286 s->low_delay= 1;
3287 avctx->pix_fmt= PIX_FMT_YUV420P;
3288
c2212338 3289 decode_init_vlc();
115329f1 3290
26165f99
MR
3291 if(avctx->extradata_size > 0 && avctx->extradata &&
3292 *(char *)avctx->extradata == 1){
4770b1b4
RT
3293 h->is_avc = 1;
3294 h->got_avcC = 0;
26165f99
MR
3295 } else {
3296 h->is_avc = 0;
4770b1b4
RT
3297 }
3298
0da71265
MN
3299 return 0;
3300}
3301
af8aa846 3302static int frame_start(H264Context *h){
0da71265
MN
3303 MpegEncContext * const s = &h->s;
3304 int i;
3305
af8aa846
MN
3306 if(MPV_frame_start(s, s->avctx) < 0)
3307 return -1;
0da71265 3308 ff_er_frame_start(s);
0da71265
MN
3309
3310 assert(s->linesize && s->uvlinesize);
3311
3312 for(i=0; i<16; i++){
3313 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 3314 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3315 }
3316 for(i=0; i<4; i++){
3317 h->block_offset[16+i]=
3318 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
3319 h->block_offset[24+16+i]=
3320 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3321 }
3322
934b0821
LM
3323 /* can't be in alloc_tables because linesize isn't known there.
3324 * FIXME: redo bipred weight to not require extra buffer? */
3325 if(!s->obmc_scratchpad)
5d18eaad
LM
3326 s->obmc_scratchpad = av_malloc(16*2*s->linesize + 8*2*s->uvlinesize);
3327
3328 /* some macroblocks will be accessed before they're available */
3329 if(FRAME_MBAFF)
3330 memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(uint8_t));
934b0821 3331
0da71265 3332// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
af8aa846 3333 return 0;
0da71265
MN
3334}
3335
53c05b1e
MN
3336static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3337 MpegEncContext * const s = &h->s;
3338 int i;
115329f1 3339
53c05b1e
MN
3340 src_y -= linesize;
3341 src_cb -= uvlinesize;
3342 src_cr -= uvlinesize;
3343
3b66c4c5 3344 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 3345 // and the line above the bottom macroblock
6ba71fc4 3346 h->left_border[0]= h->top_borders[0][s->mb_x][15];
53c05b1e
MN
3347 for(i=1; i<17; i++){
3348 h->left_border[i]= src_y[15+i* linesize];
3349 }
115329f1 3350
6ba71fc4
LLL
3351 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3352 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
53c05b1e
MN
3353
3354 if(!(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
3355 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3356 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
53c05b1e
MN
3357 for(i=1; i<9; i++){
3358 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3359 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3360 }
6ba71fc4
LLL
3361 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3362 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
53c05b1e
MN
3363 }
3364}
3365
3366static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
3367 MpegEncContext * const s = &h->s;
3368 int temp8, i;
3369 uint64_t temp64;
d89dc06a
LM
3370 int deblock_left = (s->mb_x > 0);
3371 int deblock_top = (s->mb_y > 0);
53c05b1e
MN
3372
3373 src_y -= linesize + 1;
3374 src_cb -= uvlinesize + 1;
3375 src_cr -= uvlinesize + 1;
3376
3377#define XCHG(a,b,t,xchg)\
3378t= a;\
3379if(xchg)\
3380 a= b;\
3381b= t;
d89dc06a
LM
3382
3383 if(deblock_left){
3384 for(i = !deblock_top; i<17; i++){
3385 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3386 }
3387 }
3388
3389 if(deblock_top){
6ba71fc4
LLL
3390 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3391 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
cad4368a 3392 if(s->mb_x+1 < s->mb_width){
43efd19a
LM
3393 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3394 }
53c05b1e 3395 }
53c05b1e
MN
3396
3397 if(!(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
3398 if(deblock_left){
3399 for(i = !deblock_top; i<9; i++){
3400 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3401 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3402 }
3403 }
3404 if(deblock_top){
6ba71fc4
LLL
3405 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3406 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3407 }
3408 }
3409}
3410
3411static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3412 MpegEncContext * const s = &h->s;
3413 int i;
115329f1 3414
6ba71fc4
LLL
3415 src_y -= 2 * linesize;
3416 src_cb -= 2 * uvlinesize;
3417 src_cr -= 2 * uvlinesize;
3418
3b66c4c5 3419 // There are two lines saved, the line above the the top macroblock of a pair,
6ba71fc4
LLL
3420 // and the line above the bottom macroblock
3421 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3422 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3423 for(i=2; i<34; i++){
3424 h->left_border[i]= src_y[15+i* linesize];
3425 }
115329f1 3426
6ba71fc4
LLL
3427 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3428 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3429 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3430 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3431
3432 if(!(s->flags&CODEC_FLAG_GRAY)){
3433 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3434 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3435 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3436 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3437 for(i=2; i<18; i++){
3438 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3439 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3440 }
3441 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3442 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3443 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3444 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3445 }
3446}
3447
3448static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
3449 MpegEncContext * const s = &h->s;
3450 int temp8, i;