h264: faster fill_rectangle()
[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 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
5509bffa 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0da71265
MN
18 *
19 */
115329f1 20
0da71265
MN
21/**
22 * @file h264.c
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
25 */
26
27#include "common.h"
28#include "dsputil.h"
29#include "avcodec.h"
30#include "mpegvideo.h"
31#include "h264data.h"
32#include "golomb.h"
33
e5017ab8
LA
34#include "cabac.h"
35
2848ce84 36//#undef NDEBUG
0da71265
MN
37#include <assert.h>
38
39#define interlaced_dct interlaced_dct_is_a_bad_name
40#define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42#define LUMA_DC_BLOCK_INDEX 25
43#define CHROMA_DC_BLOCK_INDEX 26
44
45#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46#define COEFF_TOKEN_VLC_BITS 8
47#define TOTAL_ZEROS_VLC_BITS 9
48#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49#define RUN_VLC_BITS 3
50#define RUN7_VLC_BITS 6
51
52#define MAX_SPS_COUNT 32
53#define MAX_PPS_COUNT 256
54
55#define MAX_MMCO_COUNT 66
56
57/**
58 * Sequence parameter set
59 */
60typedef struct SPS{
115329f1 61
0da71265
MN
62 int profile_idc;
63 int level_idc;
36940eca 64 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
0da71265
MN
65 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
66 int poc_type; ///< pic_order_cnt_type
67 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
68 int delta_pic_order_always_zero_flag;
69 int offset_for_non_ref_pic;
70 int offset_for_top_to_bottom_field;
71 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
72 int ref_frame_count; ///< num_ref_frames
a15e68de 73 int gaps_in_frame_num_allowed_flag;
0da71265
MN
74 int mb_width; ///< frame_width_in_mbs_minus1 + 1
75 int mb_height; ///< frame_height_in_mbs_minus1 + 1
76 int frame_mbs_only_flag;
77 int mb_aff; ///<mb_adaptive_frame_field_flag
78 int direct_8x8_inference_flag;
a15e68de
MN
79 int crop; ///< frame_cropping_flag
80 int crop_left; ///< frame_cropping_rect_left_offset
81 int crop_right; ///< frame_cropping_rect_right_offset
82 int crop_top; ///< frame_cropping_rect_top_offset
83 int crop_bottom; ///< frame_cropping_rect_bottom_offset
0da71265 84 int vui_parameters_present_flag;
5ff85f1d 85 AVRational sar;
ec587a50
MR
86 int timing_info_present_flag;
87 uint32_t num_units_in_tick;
88 uint32_t time_scale;
89 int fixed_frame_rate_flag;
0da71265 90 short offset_for_ref_frame[256]; //FIXME dyn aloc?
4e4d983e
LM
91 int bitstream_restriction_flag;
92 int num_reorder_frames;
239ea04c
LM
93 int scaling_matrix_present;
94 uint8_t scaling_matrix4[6][16];
95 uint8_t scaling_matrix8[2][64];
0da71265
MN
96}SPS;
97
98/**
99 * Picture parameter set
100 */
101typedef struct PPS{
102 int sps_id;
103 int cabac; ///< entropy_coding_mode_flag
104 int pic_order_present; ///< pic_order_present_flag
105 int slice_group_count; ///< num_slice_groups_minus1 + 1
106 int mb_slice_group_map_type;
107 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
108 int weighted_pred; ///< weighted_pred_flag
109 int weighted_bipred_idc;
110 int init_qp; ///< pic_init_qp_minus26 + 26
111 int init_qs; ///< pic_init_qs_minus26 + 26
112 int chroma_qp_index_offset;
113 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
114 int constrained_intra_pred; ///< constrained_intra_pred_flag
115 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
43efd19a 116 int transform_8x8_mode; ///< transform_8x8_mode_flag
239ea04c
LM
117 uint8_t scaling_matrix4[6][16];
118 uint8_t scaling_matrix8[2][64];
0da71265
MN
119}PPS;
120
121/**
122 * Memory management control operation opcode.
123 */
124typedef enum MMCOOpcode{
125 MMCO_END=0,
126 MMCO_SHORT2UNUSED,
127 MMCO_LONG2UNUSED,
128 MMCO_SHORT2LONG,
129 MMCO_SET_MAX_LONG,
115329f1 130 MMCO_RESET,
0da71265
MN
131 MMCO_LONG,
132} MMCOOpcode;
133
134/**
135 * Memory management control operation.
136 */
137typedef struct MMCO{
138 MMCOOpcode opcode;
139 int short_frame_num;
140 int long_index;
141} MMCO;
142
143/**
144 * H264Context
145 */
146typedef struct H264Context{
147 MpegEncContext s;
115329f1 148 int nal_ref_idc;
0da71265 149 int nal_unit_type;
bb270c08
DB
150#define NAL_SLICE 1
151#define NAL_DPA 2
152#define NAL_DPB 3
153#define NAL_DPC 4
154#define NAL_IDR_SLICE 5
155#define NAL_SEI 6
156#define NAL_SPS 7
157#define NAL_PPS 8
158#define NAL_AUD 9
ab470fa7
LM
159#define NAL_END_SEQUENCE 10
160#define NAL_END_STREAM 11
161#define NAL_FILLER_DATA 12
162#define NAL_SPS_EXT 13
163#define NAL_AUXILIARY_SLICE 19
0da71265 164 uint8_t *rbsp_buffer;
f038fe8b 165 unsigned int rbsp_buffer_size;
0da71265 166
4770b1b4
RT
167 /**
168 * Used to parse AVC variant of h264
169 */
170 int is_avc; ///< this flag is != 0 if codec is avc1
171 int got_avcC; ///< flag used to parse avcC data only once
172 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
173
0da71265
MN
174 int chroma_qp; //QPc
175
160d679c 176 int prev_mb_skipped; //FIXME remove (IMHO not used)
0da71265
MN
177
178 //prediction stuff
179 int chroma_pred_mode;
180 int intra16x16_pred_mode;
6ba71fc4
LLL
181
182 int top_mb_xy;
183 int left_mb_xy[2];
115329f1 184
0da71265
MN
185 int8_t intra4x4_pred_mode_cache[5*8];
186 int8_t (*intra4x4_pred_mode)[8];
187 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
43efd19a 188 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
0da71265
MN
189 void (*pred8x8 [4+3])(uint8_t *src, int stride);
190 void (*pred16x16[4+3])(uint8_t *src, int stride);
191 unsigned int topleft_samples_available;
192 unsigned int top_samples_available;
193 unsigned int topright_samples_available;
194 unsigned int left_samples_available;
6ba71fc4
LLL
195 uint8_t (*top_borders[2])[16+2*8];
196 uint8_t left_border[2*(17+2*9)];
0da71265
MN
197
198 /**
199 * non zero coeff count cache.
200 * is 64 if not available.
201 */
68b51e58 202 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
53c05b1e 203 uint8_t (*non_zero_count)[16];
0da71265
MN
204
205 /**
206 * Motion vector cache.
207 */
68b51e58
SH
208 DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
209 DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
0da71265
MN
210#define LIST_NOT_USED -1 //FIXME rename?
211#define PART_NOT_AVAILABLE -2
115329f1 212
0da71265
MN
213 /**
214 * is 1 if the specific list MV&references are set to 0,0,-2.
215 */
216 int mv_cache_clean[2];
217
6867a90b 218 /**
43efd19a
LM
219 * number of neighbors (top and/or left) that used 8x8 dct
220 */
221 int neighbor_transform_size;
222
223 /**
6867a90b
LLL
224 * block_offset[ 0..23] for frame macroblocks
225 * block_offset[24..47] for field macroblocks
226 */
227 int block_offset[2*(16+8)];
115329f1 228
a55f20bd
LM
229 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
230 uint32_t *mb2b8_xy;
650cec0c 231 int b_stride; //FIXME use s->b4_stride
0da71265
MN
232 int b8_stride;
233
8b82a956
MN
234 int halfpel_flag;
235 int thirdpel_flag;
236
da3b9756
MM
237 int unknown_svq3_flag;
238 int next_slice_index;
239
0da71265
MN
240 SPS sps_buffer[MAX_SPS_COUNT];
241 SPS sps; ///< current sps
115329f1 242
0da71265
MN
243 PPS pps_buffer[MAX_PPS_COUNT];
244 /**
245 * current pps
246 */
3b66c4c5 247 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
0da71265 248
239ea04c
LM
249 uint32_t dequant4_buffer[6][52][16];
250 uint32_t dequant8_buffer[2][52][64];
251 uint32_t (*dequant4_coeff[6])[16];
252 uint32_t (*dequant8_coeff[2])[64];
253 int dequant_coeff_pps; ///< reinit tables when pps changes
43efd19a 254
0da71265
MN
255 int slice_num;
256 uint8_t *slice_table_base;
257 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
258 int slice_type;
259 int slice_type_fixed;
115329f1 260
0da71265 261 //interlacing specific flags
6ba71fc4 262 int mb_aff_frame;
0da71265 263 int mb_field_decoding_flag;
115329f1 264
0da71265 265 int sub_mb_type[4];
115329f1 266
0da71265
MN
267 //POC stuff
268 int poc_lsb;
269 int poc_msb;
270 int delta_poc_bottom;
271 int delta_poc[2];
272 int frame_num;
273 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
274 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
275 int frame_num_offset; ///< for POC type 2
276 int prev_frame_num_offset; ///< for POC type 2
277 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
278
279 /**
280 * frame_num for frames or 2*frame_num for field pics.
281 */
282 int curr_pic_num;
115329f1 283
0da71265
MN
284 /**
285 * max_frame_num or 2*max_frame_num for field pics.
286 */
287 int max_pic_num;
288
289 //Weighted pred stuff
9f2d1b4f
LM
290 int use_weight;
291 int use_weight_chroma;
0da71265
MN
292 int luma_log2_weight_denom;
293 int chroma_log2_weight_denom;
294 int luma_weight[2][16];
295 int luma_offset[2][16];
296 int chroma_weight[2][16][2];
297 int chroma_offset[2][16][2];
9f2d1b4f 298 int implicit_weight[16][16];
115329f1 299
0da71265 300 //deblock
115329f1 301 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
980a82b7
MN
302 int slice_alpha_c0_offset;
303 int slice_beta_offset;
115329f1 304
0da71265 305 int redundant_pic_count;
115329f1 306
0da71265 307 int direct_spatial_mv_pred;
5ad984c9 308 int dist_scale_factor[16];
2f944356 309 int map_col_to_list0[2][16];
0da71265
MN
310
311 /**
312 * num_ref_idx_l0/1_active_minus1 + 1
313 */
314 int ref_count[2];// FIXME split for AFF
17107065
LM
315 Picture *short_ref[32];
316 Picture *long_ref[32];
0da71265
MN
317 Picture default_ref_list[2][32];
318 Picture ref_list[2][32]; //FIXME size?
319 Picture field_ref_list[2][32]; //FIXME size?
a4dae92b 320 Picture *delayed_pic[16]; //FIXME size?
4e4d983e 321 Picture *delayed_output_pic;
115329f1 322
0da71265
MN
323 /**
324 * memory management control operations buffer.
325 */
326 MMCO mmco[MAX_MMCO_COUNT];
327 int mmco_index;
115329f1 328
0da71265
MN
329 int long_ref_count; ///< number of actual long term references
330 int short_ref_count; ///< number of actual short term references
115329f1 331
0da71265
MN
332 //data partitioning
333 GetBitContext intra_gb;
334 GetBitContext inter_gb;
335 GetBitContext *intra_gb_ptr;
336 GetBitContext *inter_gb_ptr;
115329f1 337
68b51e58 338 DECLARE_ALIGNED_8(DCTELEM, mb[16*24]);
e5017ab8
LA
339
340 /**
341 * Cabac
342 */
343 CABACContext cabac;
43efd19a 344 uint8_t cabac_state[460];
e5017ab8
LA
345 int cabac_init_idc;
346
347 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
348 uint16_t *cbp_table;
3981c385
MN
349 int top_cbp;
350 int left_cbp;
693399ba 351 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
e5017ab8
LA
352 uint8_t *chroma_pred_mode_table;
353 int last_qscale_diff;
9e528114 354 int16_t (*mvd_table[2])[2];
68b51e58 355 DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
5ad984c9
LM
356 uint8_t *direct_table;
357 uint8_t direct_cache[5*8];
e5017ab8 358
e4b36d44
MN
359 uint8_t zigzag_scan[16];
360 uint8_t field_scan[16];
548a1c8a
LM
361 uint8_t zigzag_scan8x8[64];
362 uint8_t zigzag_scan8x8_cavlc[64];
36940eca
LM
363 const uint8_t *zigzag_scan_q0;
364 const uint8_t *field_scan_q0;
548a1c8a
LM
365 const uint8_t *zigzag_scan8x8_q0;
366 const uint8_t *zigzag_scan8x8_cavlc_q0;
115329f1 367
cdd10689 368 int x264_build;
0da71265
MN
369}H264Context;
370
371static VLC coeff_token_vlc[4];
372static VLC chroma_dc_coeff_token_vlc;
373
374static VLC total_zeros_vlc[15];
375static VLC chroma_dc_total_zeros_vlc[3];
376
377static VLC run_vlc[6];
378static VLC run7_vlc;
379
8b82a956
MN
380static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
381static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
6ba71fc4 382static 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);
8b82a956 383
70abb407 384static always_inline uint32_t pack16to32(int a, int b){
377ec888
MN
385#ifdef WORDS_BIGENDIAN
386 return (b&0xFFFF) + (a<<16);
387#else
388 return (a&0xFFFF) + (b<<16);
389#endif
390}
391
0da71265
MN
392/**
393 * fill a rectangle.
5175b937
LLL
394 * @param h height of the rectangle, should be a constant
395 * @param w width of the rectangle, should be a constant
0da71265
MN
396 * @param size the size of val (1 or 4), should be a constant
397 */
70abb407 398static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
af6e2fed 399 uint8_t *p= (uint8_t*)vp;
0da71265 400 assert(size==1 || size==4);
67a82086 401 assert(w<=4);
115329f1 402
0da71265
MN
403 w *= size;
404 stride *= size;
115329f1 405
4733abcb 406 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
d9c780a8 407 assert((stride&(w-1))==0);
67a82086
LM
408 if(w==2){
409 const uint16_t v= size==4 ? val : val*0x0101;
410 *(uint16_t*)(p + 0*stride)= v;
411 if(h==1) return;
412 *(uint16_t*)(p + 1*stride)= v;
413 if(h==2) return;
0da71265 414 *(uint16_t*)(p + 2*stride)=
67a82086
LM
415 *(uint16_t*)(p + 3*stride)= v;
416 }else if(w==4){
417 const uint32_t v= size==4 ? val : val*0x01010101;
418 *(uint32_t*)(p + 0*stride)= v;
419 if(h==1) return;
420 *(uint32_t*)(p + 1*stride)= v;
421 if(h==2) return;
0da71265 422 *(uint32_t*)(p + 2*stride)=
67a82086
LM
423 *(uint32_t*)(p + 3*stride)= v;
424 }else if(w==8){
425 //gcc can't optimize 64bit math on x86_32
426#if defined(ARCH_X86_64) || (defined(MP_WORDSIZE) && MP_WORDSIZE >= 64)
427 const uint64_t v= val*0x0100000001ULL;
428 *(uint64_t*)(p + 0*stride)= v;
429 if(h==1) return;
430 *(uint64_t*)(p + 1*stride)= v;
431 if(h==2) return;
0da71265 432 *(uint64_t*)(p + 2*stride)=
67a82086
LM
433 *(uint64_t*)(p + 3*stride)= v;
434 }else if(w==16){
435 const uint64_t v= val*0x0100000001ULL;
0da71265
MN
436 *(uint64_t*)(p + 0+0*stride)=
437 *(uint64_t*)(p + 8+0*stride)=
438 *(uint64_t*)(p + 0+1*stride)=
67a82086
LM
439 *(uint64_t*)(p + 8+1*stride)= v;
440 if(h==2) return;
0da71265
MN
441 *(uint64_t*)(p + 0+2*stride)=
442 *(uint64_t*)(p + 8+2*stride)=
443 *(uint64_t*)(p + 0+3*stride)=
67a82086
LM
444 *(uint64_t*)(p + 8+3*stride)= v;
445#else
446 *(uint32_t*)(p + 0+0*stride)=
447 *(uint32_t*)(p + 4+0*stride)= val;
448 if(h==1) return;
449 *(uint32_t*)(p + 0+1*stride)=
450 *(uint32_t*)(p + 4+1*stride)= val;
451 if(h==2) return;
452 *(uint32_t*)(p + 0+2*stride)=
453 *(uint32_t*)(p + 4+2*stride)=
454 *(uint32_t*)(p + 0+3*stride)=
455 *(uint32_t*)(p + 4+3*stride)= val;
456 }else if(w==16){
457 *(uint32_t*)(p + 0+0*stride)=
458 *(uint32_t*)(p + 4+0*stride)=
459 *(uint32_t*)(p + 8+0*stride)=
460 *(uint32_t*)(p +12+0*stride)=
461 *(uint32_t*)(p + 0+1*stride)=
462 *(uint32_t*)(p + 4+1*stride)=
463 *(uint32_t*)(p + 8+1*stride)=
464 *(uint32_t*)(p +12+1*stride)= val;
465 if(h==2) return;
466 *(uint32_t*)(p + 0+2*stride)=
467 *(uint32_t*)(p + 4+2*stride)=
468 *(uint32_t*)(p + 8+2*stride)=
469 *(uint32_t*)(p +12+2*stride)=
470 *(uint32_t*)(p + 0+3*stride)=
471 *(uint32_t*)(p + 4+3*stride)=
472 *(uint32_t*)(p + 8+3*stride)=
473 *(uint32_t*)(p +12+3*stride)= val;
474#endif
0da71265
MN
475 }else
476 assert(0);
67a82086 477 assert(h==4);
0da71265
MN
478}
479
70abb407 480static void fill_caches(H264Context *h, int mb_type, int for_deblock){
0da71265 481 MpegEncContext * const s = &h->s;
7bc9090a 482 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
483 int topleft_xy, top_xy, topright_xy, left_xy[2];
484 int topleft_type, top_type, topright_type, left_type[2];
6867a90b 485 int left_block[8];
0da71265
MN
486 int i;
487
e2e5894a
LM
488 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
489 // the actual condition is whether we're on the edge of a slice,
490 // and even then the intra and nnz parts are unnecessary.
491 if(for_deblock && h->slice_num == 1)
492 return;
493
115329f1
DB
494 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
495
6867a90b
LLL
496 top_xy = mb_xy - s->mb_stride;
497 topleft_xy = top_xy - 1;
498 topright_xy= top_xy + 1;
499 left_xy[1] = left_xy[0] = mb_xy-1;
500 left_block[0]= 0;
501 left_block[1]= 1;
502 left_block[2]= 2;
503 left_block[3]= 3;
504 left_block[4]= 7;
505 left_block[5]= 10;
506 left_block[6]= 8;
507 left_block[7]= 11;
6ba71fc4 508 if(h->mb_aff_frame){
6867a90b
LLL
509 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
510 const int top_pair_xy = pair_xy - s->mb_stride;
511 const int topleft_pair_xy = top_pair_xy - 1;
512 const int topright_pair_xy = top_pair_xy + 1;
513 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
514 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
515 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
516 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
517 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
518 const int bottom = (s->mb_y & 1);
519 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);
520 if (bottom
521 ? !curr_mb_frame_flag // bottom macroblock
522 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
523 ) {
524 top_xy -= s->mb_stride;
525 }
526 if (bottom
527 ? !curr_mb_frame_flag // bottom macroblock
528 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
529 ) {
530 topleft_xy -= s->mb_stride;
531 }
532 if (bottom
533 ? !curr_mb_frame_flag // bottom macroblock
534 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
535 ) {
536 topright_xy -= s->mb_stride;
537 }
538 if (left_mb_frame_flag != curr_mb_frame_flag) {
539 left_xy[1] = left_xy[0] = pair_xy - 1;
540 if (curr_mb_frame_flag) {
541 if (bottom) {
542 left_block[0]= 2;
543 left_block[1]= 2;
544 left_block[2]= 3;
545 left_block[3]= 3;
546 left_block[4]= 8;
547 left_block[5]= 11;
548 left_block[6]= 8;
549 left_block[7]= 11;
550 } else {
551 left_block[0]= 0;
552 left_block[1]= 0;
553 left_block[2]= 1;
554 left_block[3]= 1;
555 left_block[4]= 7;
556 left_block[5]= 10;
557 left_block[6]= 7;
558 left_block[7]= 10;
559 }
560 } else {
561 left_xy[1] += s->mb_stride;
562 //left_block[0]= 0;
563 left_block[1]= 2;
564 left_block[2]= 0;
565 left_block[3]= 2;
566 //left_block[4]= 7;
567 left_block[5]= 10;
568 left_block[6]= 7;
569 left_block[7]= 10;
570 }
571 }
0da71265
MN
572 }
573
826de46e
LLL
574 h->top_mb_xy = top_xy;
575 h->left_mb_xy[0] = left_xy[0];
576 h->left_mb_xy[1] = left_xy[1];
6ba71fc4 577 if(for_deblock){
46f2f05f
MN
578 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
579 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
580 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
581 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
582 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
583 }else{
584 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
585 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
586 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
587 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
588 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
589 }
0da71265
MN
590
591 if(IS_INTRA(mb_type)){
115329f1
DB
592 h->topleft_samples_available=
593 h->top_samples_available=
0da71265
MN
594 h->left_samples_available= 0xFFFF;
595 h->topright_samples_available= 0xEEEA;
596
597 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
598 h->topleft_samples_available= 0xB3FF;
599 h->top_samples_available= 0x33FF;
600 h->topright_samples_available= 0x26EA;
601 }
602 for(i=0; i<2; i++){
603 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
604 h->topleft_samples_available&= 0xDF5F;
605 h->left_samples_available&= 0x5F5F;
606 }
607 }
115329f1 608
0da71265
MN
609 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
610 h->topleft_samples_available&= 0x7FFF;
115329f1 611
0da71265
MN
612 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
613 h->topright_samples_available&= 0xFBFF;
115329f1 614
0da71265
MN
615 if(IS_INTRA4x4(mb_type)){
616 if(IS_INTRA4x4(top_type)){
617 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
618 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
619 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
620 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
621 }else{
622 int pred;
6fbcaaa0 623 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
0da71265 624 pred= -1;
6fbcaaa0
LLL
625 else{
626 pred= 2;
0da71265
MN
627 }
628 h->intra4x4_pred_mode_cache[4+8*0]=
629 h->intra4x4_pred_mode_cache[5+8*0]=
630 h->intra4x4_pred_mode_cache[6+8*0]=
631 h->intra4x4_pred_mode_cache[7+8*0]= pred;
632 }
633 for(i=0; i<2; i++){
634 if(IS_INTRA4x4(left_type[i])){
635 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
636 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
637 }else{
638 int pred;
6fbcaaa0 639 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
0da71265 640 pred= -1;
6fbcaaa0
LLL
641 else{
642 pred= 2;
0da71265
MN
643 }
644 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
645 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
646 }
647 }
648 }
649 }
115329f1
DB
650
651
0da71265 652/*
115329f1
DB
6530 . T T. T T T T
6541 L . .L . . . .
6552 L . .L . . . .
6563 . T TL . . . .
6574 L . .L . . . .
6585 L . .. . . . .
0da71265
MN
659*/
660//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
661 if(top_type){
6867a90b
LLL
662 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
663 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
664 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
53c05b1e 665 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
115329f1 666
6867a90b 667 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
53c05b1e 668 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
115329f1 669
6867a90b 670 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
53c05b1e 671 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
115329f1 672
0da71265 673 }else{
115329f1 674 h->non_zero_count_cache[4+8*0]=
0da71265
MN
675 h->non_zero_count_cache[5+8*0]=
676 h->non_zero_count_cache[6+8*0]=
677 h->non_zero_count_cache[7+8*0]=
115329f1 678
0da71265
MN
679 h->non_zero_count_cache[1+8*0]=
680 h->non_zero_count_cache[2+8*0]=
115329f1 681
0da71265 682 h->non_zero_count_cache[1+8*3]=
3981c385 683 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
115329f1 684
0da71265 685 }
826de46e 686
6867a90b
LLL
687 for (i=0; i<2; i++) {
688 if(left_type[i]){
689 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
690 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
691 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
692 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
6867a90b 693 }else{
115329f1
DB
694 h->non_zero_count_cache[3+8*1 + 2*8*i]=
695 h->non_zero_count_cache[3+8*2 + 2*8*i]=
696 h->non_zero_count_cache[0+8*1 + 8*i]=
6867a90b 697 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
826de46e
LLL
698 }
699 }
700
701 if( h->pps.cabac ) {
702 // top_cbp
703 if(top_type) {
704 h->top_cbp = h->cbp_table[top_xy];
705 } else if(IS_INTRA(mb_type)) {
706 h->top_cbp = 0x1C0;
707 } else {
708 h->top_cbp = 0;
709 }
710 // left_cbp
711 if (left_type[0]) {
712 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
713 } else if(IS_INTRA(mb_type)) {
714 h->left_cbp = 0x1C0;
715 } else {
716 h->left_cbp = 0;
717 }
718 if (left_type[0]) {
719 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
720 }
721 if (left_type[1]) {
722 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
6867a90b 723 }
0da71265 724 }
6867a90b 725
0da71265 726#if 1
5ad984c9 727 //FIXME direct mb can skip much of this
e2e5894a 728 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
0da71265 729 int list;
e2e5894a
LM
730 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
731 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
0da71265
MN
732 /*if(!h->mv_cache_clean[list]){
733 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
734 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
735 h->mv_cache_clean[list]= 1;
736 }*/
5ad984c9 737 continue;
0da71265
MN
738 }
739 h->mv_cache_clean[list]= 0;
115329f1 740
53b19144 741 if(USES_LIST(top_type, list)){
0da71265
MN
742 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
743 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
744 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
745 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
746 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
747 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
748 h->ref_cache[list][scan8[0] + 0 - 1*8]=
749 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
750 h->ref_cache[list][scan8[0] + 2 - 1*8]=
751 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
752 }else{
115329f1
DB
753 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
754 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
755 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
0da71265
MN
756 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
757 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
758 }
759
0da71265 760 //FIXME unify cleanup or sth
53b19144 761 if(USES_LIST(left_type[0], list)){
0da71265
MN
762 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
763 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
764 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
765 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
115329f1 766 h->ref_cache[list][scan8[0] - 1 + 0*8]=
0da71265
MN
767 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
768 }else{
769 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
770 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
771 h->ref_cache[list][scan8[0] - 1 + 0*8]=
772 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
773 }
115329f1 774
53b19144 775 if(USES_LIST(left_type[1], list)){
0da71265
MN
776 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
777 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
778 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
779 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
115329f1 780 h->ref_cache[list][scan8[0] - 1 + 2*8]=
0da71265
MN
781 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
782 }else{
783 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
784 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
785 h->ref_cache[list][scan8[0] - 1 + 2*8]=
786 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
d9c780a8 787 assert((!left_type[0]) == (!left_type[1]));
0da71265
MN
788 }
789
e2e5894a 790 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
46f2f05f
MN
791 continue;
792
53b19144 793 if(USES_LIST(topleft_type, list)){
e2e5894a
LM
794 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
795 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
796 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
797 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
798 }else{
799 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
800 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
801 }
115329f1 802
53b19144 803 if(USES_LIST(topright_type, list)){
e2e5894a
LM
804 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
805 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
806 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
807 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
808 }else{
809 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
810 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
811 }
e2e5894a 812
115329f1
DB
813
814 h->ref_cache[list][scan8[5 ]+1] =
815 h->ref_cache[list][scan8[7 ]+1] =
3b66c4c5 816 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
115329f1 817 h->ref_cache[list][scan8[4 ]] =
0da71265
MN
818 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
819 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
820 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
3b66c4c5 821 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
0da71265
MN
822 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
823 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
9e528114
LA
824
825 if( h->pps.cabac ) {
826 /* XXX beurk, Load mvd */
53b19144 827 if(USES_LIST(topleft_type, list)){
9e528114
LA
828 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
829 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
830 }else{
831 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
832 }
833
53b19144 834 if(USES_LIST(top_type, list)){
9e528114
LA
835 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
836 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
837 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
838 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
839 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
840 }else{
115329f1
DB
841 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
842 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
843 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
9e528114
LA
844 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
845 }
53b19144 846 if(USES_LIST(left_type[0], list)){
9e528114
LA
847 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
848 *(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]];
849 *(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]];
850 }else{
851 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
852 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
853 }
53b19144 854 if(USES_LIST(left_type[1], list)){
9e528114
LA
855 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
856 *(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]];
857 *(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]];
858 }else{
859 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
860 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
861 }
862 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
863 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
3b66c4c5 864 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
9e528114
LA
865 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
866 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
5ad984c9
LM
867
868 if(h->slice_type == B_TYPE){
869 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
870
871 if(IS_DIRECT(top_type)){
872 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
873 }else if(IS_8X8(top_type)){
874 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
875 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
876 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
877 }else{
878 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
879 }
115329f1 880
5ad984c9
LM
881 //FIXME interlacing
882 if(IS_DIRECT(left_type[0])){
883 h->direct_cache[scan8[0] - 1 + 0*8]=
884 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
885 }else if(IS_8X8(left_type[0])){
886 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
887 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
888 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
889 }else{
890 h->direct_cache[scan8[0] - 1 + 0*8]=
891 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
892 }
893 }
9e528114 894 }
0da71265 895 }
0da71265
MN
896 }
897#endif
43efd19a
LM
898
899 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
0da71265
MN
900}
901
902static inline void write_back_intra_pred_mode(H264Context *h){
903 MpegEncContext * const s = &h->s;
7bc9090a 904 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
905
906 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
907 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
908 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
909 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
910 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
911 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
912 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
913}
914
915/**
916 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
917 */
918static inline int check_intra4x4_pred_mode(H264Context *h){
919 MpegEncContext * const s = &h->s;
920 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
921 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
922 int i;
115329f1 923
0da71265
MN
924 if(!(h->top_samples_available&0x8000)){
925 for(i=0; i<4; i++){
926 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
927 if(status<0){
9b879566 928 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
929 return -1;
930 } else if(status){
931 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
932 }
933 }
934 }
115329f1 935
0da71265
MN
936 if(!(h->left_samples_available&0x8000)){
937 for(i=0; i<4; i++){
938 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
939 if(status<0){
9b879566 940 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
941 return -1;
942 } else if(status){
943 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
944 }
945 }
946 }
947
948 return 0;
949} //FIXME cleanup like next
950
951/**
952 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
953 */
954static inline int check_intra_pred_mode(H264Context *h, int mode){
955 MpegEncContext * const s = &h->s;
956 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
957 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 958
5175b937
LLL
959 if(mode < 0 || mode > 6) {
960 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 961 return -1;
5175b937 962 }
115329f1 963
0da71265
MN
964 if(!(h->top_samples_available&0x8000)){
965 mode= top[ mode ];
966 if(mode<0){
9b879566 967 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
968 return -1;
969 }
970 }
115329f1 971
0da71265
MN
972 if(!(h->left_samples_available&0x8000)){
973 mode= left[ mode ];
974 if(mode<0){
9b879566 975 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 976 return -1;
115329f1 977 }
0da71265
MN
978 }
979
980 return mode;
981}
982
983/**
984 * gets the predicted intra4x4 prediction mode.
985 */
986static inline int pred_intra_mode(H264Context *h, int n){
987 const int index8= scan8[n];
988 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
989 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
990 const int min= FFMIN(left, top);
991
95c26348 992 tprintf("mode:%d %d min:%d\n", left ,top, min);
0da71265
MN
993
994 if(min<0) return DC_PRED;
995 else return min;
996}
997
998static inline void write_back_non_zero_count(H264Context *h){
999 MpegEncContext * const s = &h->s;
7bc9090a 1000 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 1001
6867a90b
LLL
1002 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1003 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1004 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
53c05b1e 1005 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
6867a90b
LLL
1006 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1007 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1008 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
115329f1 1009
6867a90b 1010 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
53c05b1e 1011 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
6867a90b 1012 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
53c05b1e 1013
6867a90b 1014 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
53c05b1e 1015 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
6867a90b 1016 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
0da71265
MN
1017}
1018
1019/**
1020 * gets the predicted number of non zero coefficients.
1021 * @param n block index
1022 */
1023static inline int pred_non_zero_count(H264Context *h, int n){
1024 const int index8= scan8[n];
1025 const int left= h->non_zero_count_cache[index8 - 1];
1026 const int top = h->non_zero_count_cache[index8 - 8];
1027 int i= left + top;
115329f1 1028
0da71265
MN
1029 if(i<64) i= (i+1)>>1;
1030
95c26348 1031 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
0da71265
MN
1032
1033 return i&31;
1034}
1035
1924f3ce
MN
1036static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1037 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1038
1039 if(topright_ref != PART_NOT_AVAILABLE){
1040 *C= h->mv_cache[list][ i - 8 + part_width ];
1041 return topright_ref;
1042 }else{
95c26348
MN
1043 tprintf("topright MV not available\n");
1044
1924f3ce
MN
1045 *C= h->mv_cache[list][ i - 8 - 1 ];
1046 return h->ref_cache[list][ i - 8 - 1 ];
1047 }
1048}
1049
0da71265
MN
1050/**
1051 * gets the predicted MV.
1052 * @param n the block index
1053 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1054 * @param mx the x component of the predicted motion vector
1055 * @param my the y component of the predicted motion vector
1056 */
1057static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
0da71265
MN
1058 const int index8= scan8[n];
1059 const int top_ref= h->ref_cache[list][ index8 - 8 ];
0da71265
MN
1060 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1061 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1062 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1924f3ce
MN
1063 const int16_t * C;
1064 int diagonal_ref, match_count;
1065
0da71265 1066 assert(part_width==1 || part_width==2 || part_width==4);
1924f3ce 1067
0da71265 1068/* mv_cache
115329f1 1069 B . . A T T T T
0da71265
MN
1070 U . . L . . , .
1071 U . . L . . . .
1072 U . . L . . , .
1073 . . . L . . . .
1074*/
1924f3ce
MN
1075
1076 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1077 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
827c91bf 1078 tprintf("pred_motion match_count=%d\n", match_count);
1924f3ce
MN
1079 if(match_count > 1){ //most common
1080 *mx= mid_pred(A[0], B[0], C[0]);
1081 *my= mid_pred(A[1], B[1], C[1]);
1082 }else if(match_count==1){
1083 if(left_ref==ref){
1084 *mx= A[0];
115329f1 1085 *my= A[1];
1924f3ce
MN
1086 }else if(top_ref==ref){
1087 *mx= B[0];
115329f1 1088 *my= B[1];
0da71265 1089 }else{
1924f3ce 1090 *mx= C[0];
115329f1 1091 *my= C[1];
0da71265
MN
1092 }
1093 }else{
1924f3ce 1094 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
0da71265 1095 *mx= A[0];
115329f1 1096 *my= A[1];
0da71265 1097 }else{
1924f3ce
MN
1098 *mx= mid_pred(A[0], B[0], C[0]);
1099 *my= mid_pred(A[1], B[1], C[1]);
0da71265 1100 }
0da71265 1101 }
115329f1 1102
af6e2fed 1103 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
1104}
1105
1106/**
1107 * gets the directionally predicted 16x8 MV.
1108 * @param n the block index
1109 * @param mx the x component of the predicted motion vector
1110 * @param my the y component of the predicted motion vector
1111 */
1112static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1113 if(n==0){
1114 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1115 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1116
579448a0 1117 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 1118
0da71265
MN
1119 if(top_ref == ref){
1120 *mx= B[0];
1121 *my= B[1];
1122 return;
1123 }
1124 }else{
1125 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1126 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
115329f1 1127
579448a0 1128 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
1129
1130 if(left_ref == ref){
1131 *mx= A[0];
1132 *my= A[1];
1133 return;
1134 }
1135 }
1136
1137 //RARE
1138 pred_motion(h, n, 4, list, ref, mx, my);
1139}
1140
1141/**
1142 * gets the directionally predicted 8x16 MV.
1143 * @param n the block index
1144 * @param mx the x component of the predicted motion vector
1145 * @param my the y component of the predicted motion vector
1146 */
1147static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1148 if(n==0){
1149 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1150 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
115329f1 1151
579448a0 1152 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
1153
1154 if(left_ref == ref){
1155 *mx= A[0];
1156 *my= A[1];
1157 return;
1158 }
1159 }else{
1924f3ce
MN
1160 const int16_t * C;
1161 int diagonal_ref;
1162
1163 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
115329f1 1164
579448a0 1165 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 1166
115329f1 1167 if(diagonal_ref == ref){
0da71265
MN
1168 *mx= C[0];
1169 *my= C[1];
1170 return;
1171 }
0da71265
MN
1172 }
1173
1174 //RARE
1175 pred_motion(h, n, 2, list, ref, mx, my);
1176}
1177
1178static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
0da71265
MN
1179 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1180 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1181
5175b937 1182 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
0da71265
MN
1183
1184 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1185 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1186 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
115329f1 1187
0da71265
MN
1188 *mx = *my = 0;
1189 return;
1190 }
115329f1 1191
0da71265
MN
1192 pred_motion(h, 0, 4, 0, 0, mx, my);
1193
1194 return;
1195}
1196
5ad984c9
LM
1197static inline void direct_dist_scale_factor(H264Context * const h){
1198 const int poc = h->s.current_picture_ptr->poc;
1199 const int poc1 = h->ref_list[1][0].poc;
1200 int i;
1201 for(i=0; i<h->ref_count[0]; i++){
1202 int poc0 = h->ref_list[0][i].poc;
1203 int td = clip(poc1 - poc0, -128, 127);
1204 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1205 h->dist_scale_factor[i] = 256;
1206 }else{
1207 int tb = clip(poc - poc0, -128, 127);
1208 int tx = (16384 + (ABS(td) >> 1)) / td;
1209 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1210 }
1211 }
1212}
2f944356
LM
1213static inline void direct_ref_list_init(H264Context * const h){
1214 MpegEncContext * const s = &h->s;
1215 Picture * const ref1 = &h->ref_list[1][0];
1216 Picture * const cur = s->current_picture_ptr;
1217 int list, i, j;
1218 if(cur->pict_type == I_TYPE)
1219 cur->ref_count[0] = 0;
1220 if(cur->pict_type != B_TYPE)
1221 cur->ref_count[1] = 0;
1222 for(list=0; list<2; list++){
1223 cur->ref_count[list] = h->ref_count[list];
1224 for(j=0; j<h->ref_count[list]; j++)
1225 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1226 }
1227 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1228 return;
1229 for(list=0; list<2; list++){
1230 for(i=0; i<ref1->ref_count[list]; i++){
1231 const int poc = ref1->ref_poc[list][i];
171c4076 1232 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
2f944356
LM
1233 for(j=0; j<h->ref_count[list]; j++)
1234 if(h->ref_list[list][j].poc == poc){
1235 h->map_col_to_list0[list][i] = j;
1236 break;
1237 }
1238 }
1239 }
1240}
5ad984c9
LM
1241
1242static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1243 MpegEncContext * const s = &h->s;
1244 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1245 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1246 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1247 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1248 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
4866bd2b 1249 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
5ad984c9 1250 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
2f944356 1251 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
5ad984c9
LM
1252 const int is_b8x8 = IS_8X8(*mb_type);
1253 int sub_mb_type;
1254 int i8, i4;
1255
1256 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1257 /* FIXME save sub mb types from previous frames (or derive from MVs)
1258 * so we know exactly what block size to use */
1259 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
3622988f 1260 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1261 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1262 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1263 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1264 }else{
1265 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
3622988f 1266 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1267 }
1268 if(!is_b8x8)
1269 *mb_type |= MB_TYPE_DIRECT2;
1270
827c91bf 1271 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 1272
5ad984c9
LM
1273 if(h->direct_spatial_mv_pred){
1274 int ref[2];
1275 int mv[2][2];
1276 int list;
1277
1278 /* ref = min(neighbors) */
1279 for(list=0; list<2; list++){
1280 int refa = h->ref_cache[list][scan8[0] - 1];
1281 int refb = h->ref_cache[list][scan8[0] - 8];
1282 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1283 if(refc == -2)
1284 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1285 ref[list] = refa;
1286 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1287 ref[list] = refb;
1288 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1289 ref[list] = refc;
1290 if(ref[list] < 0)
1291 ref[list] = -1;
1292 }
1293
1294 if(ref[0] < 0 && ref[1] < 0){
1295 ref[0] = ref[1] = 0;
1296 mv[0][0] = mv[0][1] =
1297 mv[1][0] = mv[1][1] = 0;
1298 }else{
1299 for(list=0; list<2; list++){
1300 if(ref[list] >= 0)
1301 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1302 else
1303 mv[list][0] = mv[list][1] = 0;
1304 }
1305 }
1306
1307 if(ref[1] < 0){
1308 *mb_type &= ~MB_TYPE_P0L1;
1309 sub_mb_type &= ~MB_TYPE_P0L1;
1310 }else if(ref[0] < 0){
1311 *mb_type &= ~MB_TYPE_P0L0;
1312 sub_mb_type &= ~MB_TYPE_P0L0;
1313 }
1314
1315 if(IS_16X16(*mb_type)){
cec93959
LM
1316 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1317 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
115329f1 1318 if(!IS_INTRA(mb_type_col)
bf4e3bd2
MR
1319 && ( (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1320 || (l1ref0[0] < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1321 && (h->x264_build>33 || !h->x264_build)))){
5ad984c9
LM
1322 if(ref[0] > 0)
1323 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1324 else
1325 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1326 if(ref[1] > 0)
1327 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1328 else
1329 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1330 }else{
1331 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1332 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1333 }
1334 }else{
1335 for(i8=0; i8<4; i8++){
1336 const int x8 = i8&1;
1337 const int y8 = i8>>1;
115329f1 1338
5ad984c9
LM
1339 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1340 continue;
1341 h->sub_mb_type[i8] = sub_mb_type;
115329f1 1342
5ad984c9
LM
1343 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1344 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
cec93959
LM
1345 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1346 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
115329f1 1347
5ad984c9 1348 /* col_zero_flag */
115329f1
DB
1349 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1350 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
bf4e3bd2 1351 && (h->x264_build>33 || !h->x264_build)))){
4866bd2b 1352 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
f1f17e54
LM
1353 if(IS_SUB_8X8(sub_mb_type)){
1354 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1355 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1356 if(ref[0] == 0)
1357 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1358 if(ref[1] == 0)
1359 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1360 }
1361 }else
5ad984c9 1362 for(i4=0; i4<4; i4++){
4866bd2b 1363 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9
LM
1364 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1365 if(ref[0] == 0)
1366 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1367 if(ref[1] == 0)
1368 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1369 }
1370 }
1371 }
1372 }
1373 }
1374 }else{ /* direct temporal mv pred */
5ad984c9
LM
1375 if(IS_16X16(*mb_type)){
1376 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1377 if(IS_INTRA(mb_type_col)){
1378 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1379 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1380 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1381 }else{
2f944356
LM
1382 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1383 : h->map_col_to_list0[1][l1ref1[0]];
5ad984c9 1384 const int dist_scale_factor = h->dist_scale_factor[ref0];
8583bef8 1385 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
5ad984c9
LM
1386 int mv_l0[2];
1387 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1388 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1389 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1390 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1391 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1392 }
1393 }else{
1394 for(i8=0; i8<4; i8++){
1395 const int x8 = i8&1;
1396 const int y8 = i8>>1;
1397 int ref0, dist_scale_factor;
bf4e3bd2 1398 const int16_t (*l1mv)[2]= l1mv0;
8583bef8 1399
5ad984c9
LM
1400 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1401 continue;
1402 h->sub_mb_type[i8] = sub_mb_type;
1403 if(IS_INTRA(mb_type_col)){
1404 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1405 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1406 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1407 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1408 continue;
1409 }
115329f1 1410
5ad984c9 1411 ref0 = l1ref0[x8 + y8*h->b8_stride];
2f944356
LM
1412 if(ref0 >= 0)
1413 ref0 = h->map_col_to_list0[0][ref0];
8583bef8 1414 else{
2f944356 1415 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
8583bef8
MN
1416 l1mv= l1mv1;
1417 }
5ad984c9 1418 dist_scale_factor = h->dist_scale_factor[ref0];
115329f1 1419
5ad984c9
LM
1420 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1421 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
f1f17e54
LM
1422 if(IS_SUB_8X8(sub_mb_type)){
1423 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1424 int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
1425 int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
1426 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1427 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1428 }else
5ad984c9 1429 for(i4=0; i4<4; i4++){
8583bef8 1430 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9
LM
1431 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1432 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1433 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1434 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1435 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1436 }
1437 }
1438 }
1439 }
1440}
1441
0da71265
MN
1442static inline void write_back_motion(H264Context *h, int mb_type){
1443 MpegEncContext * const s = &h->s;
0da71265
MN
1444 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1445 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1446 int list;
1447
2ea39252
LM
1448 if(!USES_LIST(mb_type, 0))
1449 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1450
0da71265
MN
1451 for(list=0; list<2; list++){
1452 int y;
53b19144 1453 if(!USES_LIST(mb_type, list))
5ad984c9 1454 continue;
115329f1 1455
0da71265
MN
1456 for(y=0; y<4; y++){
1457 *(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];
1458 *(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];
1459 }
9e528114
LA
1460 if( h->pps.cabac ) {
1461 for(y=0; y<4; y++){
1462 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1463 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1464 }
1465 }
53b19144
LM
1466
1467 {
1468 uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1469 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1470 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1471 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1472 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
0da71265
MN
1473 }
1474 }
115329f1 1475
5ad984c9
LM
1476 if(h->slice_type == B_TYPE && h->pps.cabac){
1477 if(IS_8X8(mb_type)){
53b19144
LM
1478 uint8_t *direct_table = &h->direct_table[b8_xy];
1479 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1480 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1481 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
5ad984c9
LM
1482 }
1483 }
0da71265
MN
1484}
1485
1486/**
1487 * Decodes a network abstraction layer unit.
1488 * @param consumed is the number of bytes used as input
1489 * @param length is the length of the array
3b66c4c5 1490 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
115329f1 1491 * @returns decoded bytes, might be src+1 if no escapes
0da71265
MN
1492 */
1493static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1494 int i, si, di;
1495 uint8_t *dst;
1496
bb270c08 1497// src[0]&0x80; //forbidden bit
0da71265
MN
1498 h->nal_ref_idc= src[0]>>5;
1499 h->nal_unit_type= src[0]&0x1F;
1500
1501 src++; length--;
115329f1 1502#if 0
0da71265
MN
1503 for(i=0; i<length; i++)
1504 printf("%2X ", src[i]);
1505#endif
1506 for(i=0; i+1<length; i+=2){
1507 if(src[i]) continue;
1508 if(i>0 && src[i-1]==0) i--;
1509 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1510 if(src[i+2]!=3){
1511 /* startcode, so we must be past the end */
1512 length=i;
1513 }
1514 break;
1515 }
1516 }
1517
1518 if(i>=length-1){ //no escaped 0
1519 *dst_length= length;
1520 *consumed= length+1; //+1 for the header
115329f1 1521 return src;
0da71265
MN
1522 }
1523
1524 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1525 dst= h->rbsp_buffer;
1526
3b66c4c5 1527//printf("decoding esc\n");
0da71265 1528 si=di=0;
115329f1 1529 while(si<length){
0da71265
MN
1530 //remove escapes (very rare 1:2^22)
1531 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1532 if(src[si+2]==3){ //escape
1533 dst[di++]= 0;
1534 dst[di++]= 0;
1535 si+=3;
c8470cc1 1536 continue;
0da71265
MN
1537 }else //next start code
1538 break;
1539 }
1540
1541 dst[di++]= src[si++];
1542 }
1543
1544 *dst_length= di;
1545 *consumed= si + 1;//+1 for the header
1546//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1547 return dst;
1548}
1549
e5017ab8 1550#if 0
0da71265
MN
1551/**
1552 * @param src the data which should be escaped
1553 * @param dst the target buffer, dst+1 == src is allowed as a special case
1554 * @param length the length of the src data
1555 * @param dst_length the length of the dst array
1556 * @returns length of escaped data in bytes or -1 if an error occured
1557 */
1558static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1559 int i, escape_count, si, di;
1560 uint8_t *temp;
115329f1 1561
0da71265
MN
1562 assert(length>=0);
1563 assert(dst_length>0);
115329f1 1564
0da71265
MN
1565 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1566
1567 if(length==0) return 1;
1568
1569 escape_count= 0;
1570 for(i=0; i<length; i+=2){
1571 if(src[i]) continue;
115329f1 1572 if(i>0 && src[i-1]==0)
0da71265
MN
1573 i--;
1574 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1575 escape_count++;
1576 i+=2;
1577 }
1578 }
115329f1
DB
1579
1580 if(escape_count==0){
0da71265
MN
1581 if(dst+1 != src)
1582 memcpy(dst+1, src, length);
1583 return length + 1;
1584 }
115329f1 1585
0da71265
MN
1586 if(length + escape_count + 1> dst_length)
1587 return -1;
1588
1589 //this should be damn rare (hopefully)
1590
1591 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1592 temp= h->rbsp_buffer;
1593//printf("encoding esc\n");
115329f1 1594
0da71265
MN
1595 si= 0;
1596 di= 0;
1597 while(si < length){
1598 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1599 temp[di++]= 0; si++;
1600 temp[di++]= 0; si++;
115329f1 1601 temp[di++]= 3;
0da71265
MN
1602 temp[di++]= src[si++];
1603 }
1604 else
1605 temp[di++]= src[si++];
1606 }
1607 memcpy(dst+1, temp, length+escape_count);
115329f1 1608
0da71265 1609 assert(di == length+escape_count);
115329f1 1610
0da71265
MN
1611 return di + 1;
1612}
1613
1614/**
1615 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1616 */
1617static void encode_rbsp_trailing(PutBitContext *pb){
1618 int length;
1619 put_bits(pb, 1, 1);
fe455f33 1620 length= (-put_bits_count(pb))&7;
0da71265
MN
1621 if(length) put_bits(pb, length, 0);
1622}
e5017ab8 1623#endif
0da71265
MN
1624
1625/**
1626 * identifies the exact end of the bitstream
1627 * @return the length of the trailing, or 0 if damaged
1628 */
1629static int decode_rbsp_trailing(uint8_t *src){
1630 int v= *src;
1631 int r;
1632
95c26348 1633 tprintf("rbsp trailing %X\n", v);
0da71265
MN
1634
1635 for(r=1; r<9; r++){
1636 if(v&1) return r;
1637 v>>=1;
1638 }
1639 return 0;
1640}
1641
1642/**
1643 * idct tranforms the 16 dc values and dequantize them.
1644 * @param qp quantization parameter
1645 */
239ea04c 1646static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1647#define stride 16
1648 int i;
1649 int temp[16]; //FIXME check if this is a good idea
1650 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1651 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1652
1653//memset(block, 64, 2*256);
1654//return;
1655 for(i=0; i<4; i++){
1656 const int offset= y_offset[i];
1657 const int z0= block[offset+stride*0] + block[offset+stride*4];
1658 const int z1= block[offset+stride*0] - block[offset+stride*4];
1659 const int z2= block[offset+stride*1] - block[offset+stride*5];
1660 const int z3= block[offset+stride*1] + block[offset+stride*5];
1661
1662 temp[4*i+0]= z0+z3;
1663 temp[4*i+1]= z1+z2;
1664 temp[4*i+2]= z1-z2;
1665 temp[4*i+3]= z0-z3;
1666 }
1667
1668 for(i=0; i<4; i++){
1669 const int offset= x_offset[i];
1670 const int z0= temp[4*0+i] + temp[4*2+i];
1671 const int z1= temp[4*0+i] - temp[4*2+i];
1672 const int z2= temp[4*1+i] - temp[4*3+i];
1673 const int z3= temp[4*1+i] + temp[4*3+i];
1674
239ea04c
LM
1675 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1676 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1677 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1678 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
0da71265
MN
1679 }
1680}
1681
e5017ab8 1682#if 0
0da71265
MN
1683/**
1684 * dct tranforms the 16 dc values.
1685 * @param qp quantization parameter ??? FIXME
1686 */
1687static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1688// const int qmul= dequant_coeff[qp][0];
1689 int i;
1690 int temp[16]; //FIXME check if this is a good idea
1691 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1692 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1693
1694 for(i=0; i<4; i++){
1695 const int offset= y_offset[i];
1696 const int z0= block[offset+stride*0] + block[offset+stride*4];
1697 const int z1= block[offset+stride*0] - block[offset+stride*4];
1698 const int z2= block[offset+stride*1] - block[offset+stride*5];
1699 const int z3= block[offset+stride*1] + block[offset+stride*5];
1700
1701 temp[4*i+0]= z0+z3;
1702 temp[4*i+1]= z1+z2;
1703 temp[4*i+2]= z1-z2;
1704 temp[4*i+3]= z0-z3;
1705 }
1706
1707 for(i=0; i<4; i++){
1708 const int offset= x_offset[i];
1709 const int z0= temp[4*0+i] + temp[4*2+i];
1710 const int z1= temp[4*0+i] - temp[4*2+i];
1711 const int z2= temp[4*1+i] - temp[4*3+i];
1712 const int z3= temp[4*1+i] + temp[4*3+i];
1713
1714 block[stride*0 +offset]= (z0 + z3)>>1;
1715 block[stride*2 +offset]= (z1 + z2)>>1;
1716 block[stride*8 +offset]= (z1 - z2)>>1;
1717 block[stride*10+offset]= (z0 - z3)>>1;
1718 }
1719}
e5017ab8
LA
1720#endif
1721
0da71265
MN
1722#undef xStride
1723#undef stride
1724
239ea04c 1725static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1726 const int stride= 16*2;
1727 const int xStride= 16;
1728 int a,b,c,d,e;
1729
1730 a= block[stride*0 + xStride*0];
1731 b= block[stride*0 + xStride*1];
1732 c= block[stride*1 + xStride*0];
1733 d= block[stride*1 + xStride*1];
1734
1735 e= a-b;
1736 a= a+b;
1737 b= c-d;
1738 c= c+d;
1739
239ea04c
LM
1740 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1741 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1742 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1743 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
0da71265
MN
1744}
1745
e5017ab8 1746#if 0
0da71265
MN
1747static void chroma_dc_dct_c(DCTELEM *block){
1748 const int stride= 16*2;
1749 const int xStride= 16;
1750 int a,b,c,d,e;
1751
1752 a= block[stride*0 + xStride*0];
1753 b= block[stride*0 + xStride*1];
1754 c= block[stride*1 + xStride*0];
1755 d= block[stride*1 + xStride*1];
1756
1757 e= a-b;
1758 a= a+b;
1759 b= c-d;
1760 c= c+d;
1761
1762 block[stride*0 + xStride*0]= (a+c);
1763 block[stride*0 + xStride*1]= (e+b);
1764 block[stride*1 + xStride*0]= (a-c);
1765 block[stride*1 + xStride*1]= (e-b);
1766}
e5017ab8 1767#endif
0da71265
MN
1768
1769/**
1770 * gets the chroma qp.
1771 */
6ba71fc4 1772static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
115329f1 1773
6ba71fc4 1774 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
0da71265
MN
1775}
1776
1777
e5017ab8 1778#if 0
0da71265
MN
1779static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1780 int i;
1781 //FIXME try int temp instead of block
115329f1 1782
0da71265
MN
1783 for(i=0; i<4; i++){
1784 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1785 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1786 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1787 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1788 const int z0= d0 + d3;
1789 const int z3= d0 - d3;
1790 const int z1= d1 + d2;
1791 const int z2= d1 - d2;
115329f1 1792
0da71265
MN
1793 block[0 + 4*i]= z0 + z1;
1794 block[1 + 4*i]= 2*z3 + z2;
1795 block[2 + 4*i]= z0 - z1;
1796 block[3 + 4*i]= z3 - 2*z2;
115329f1 1797 }
0da71265
MN
1798
1799 for(i=0; i<4; i++){
1800 const int z0= block[0*4 + i] + block[3*4 + i];
1801 const int z3= block[0*4 + i] - block[3*4 + i];
1802 const int z1= block[1*4 + i] + block[2*4 + i];
1803 const int z2= block[1*4 + i] - block[2*4 + i];
115329f1 1804
0da71265
MN
1805 block[0*4 + i]= z0 + z1;
1806 block[1*4 + i]= 2*z3 + z2;
1807 block[2*4 + i]= z0 - z1;
1808 block[3*4 + i]= z3 - 2*z2;
1809 }
1810}
e5017ab8 1811#endif
0da71265 1812
3b66c4c5 1813//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
0da71265
MN
1814//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1815static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1816 int i;
1817 const int * const quant_table= quant_coeff[qscale];
1818 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1819 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1820 const unsigned int threshold2= (threshold1<<1);
1821 int last_non_zero;
1822
1823 if(seperate_dc){
1824 if(qscale<=18){
1825 //avoid overflows
1826 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1827 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1828 const unsigned int dc_threshold2= (dc_threshold1<<1);
1829
1830 int level= block[0]*quant_coeff[qscale+18][0];
1831 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1832 if(level>0){
1833 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1834 block[0]= level;
1835 }else{
1836 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1837 block[0]= -level;
1838 }
1839// last_non_zero = i;
1840 }else{
1841 block[0]=0;
1842 }
1843 }else{
1844 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1845 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1846 const unsigned int dc_threshold2= (dc_threshold1<<1);
1847
1848 int level= block[0]*quant_table[0];
1849 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1850 if(level>0){
1851 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1852 block[0]= level;
1853 }else{
1854 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1855 block[0]= -level;
1856 }
1857// last_non_zero = i;
1858 }else{
1859 block[0]=0;
1860 }
1861 }
1862 last_non_zero= 0;
1863 i=1;
1864 }else{
1865 last_non_zero= -1;
1866 i=0;
1867 }
1868
1869 for(; i<16; i++){
1870 const int j= scantable[i];
1871 int level= block[j]*quant_table[j];
1872
1873// if( bias+level >= (1<<(QMAT_SHIFT - 3))
1874// || bias-level >= (1<<(QMAT_SHIFT - 3))){
1875 if(((unsigned)(level+threshold1))>threshold2){
1876 if(level>0){
1877 level= (bias + level)>>QUANT_SHIFT;
1878 block[j]= level;
1879 }else{
1880 level= (bias - level)>>QUANT_SHIFT;
1881 block[j]= -level;
1882 }
1883 last_non_zero = i;
1884 }else{
1885 block[j]=0;
1886 }
1887 }
1888
1889 return last_non_zero;
1890}
1891
1892static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1893 const uint32_t a= ((uint32_t*)(src-stride))[0];
1894 ((uint32_t*)(src+0*stride))[0]= a;
1895 ((uint32_t*)(src+1*stride))[0]= a;
1896 ((uint32_t*)(src+2*stride))[0]= a;
1897 ((uint32_t*)(src+3*stride))[0]= a;
1898}
1899
1900static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1901 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1902 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1903 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1904 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1905}
1906
1907static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1908 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1909 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
115329f1
DB
1910
1911 ((uint32_t*)(src+0*stride))[0]=
1912 ((uint32_t*)(src+1*stride))[0]=
1913 ((uint32_t*)(src+2*stride))[0]=
1914 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1915}
1916
1917static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1918 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
115329f1
DB
1919
1920 ((uint32_t*)(src+0*stride))[0]=
1921 ((uint32_t*)(src+1*stride))[0]=
1922 ((uint32_t*)(src+2*stride))[0]=
1923 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1924}
1925
1926static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1927 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
115329f1
DB
1928
1929 ((uint32_t*)(src+0*stride))[0]=
1930 ((uint32_t*)(src+1*stride))[0]=
1931 ((uint32_t*)(src+2*stride))[0]=
1932 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1933}
1934
1935static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1936 ((uint32_t*)(src+0*stride))[0]=
1937 ((uint32_t*)(src+1*stride))[0]=
1938 ((uint32_t*)(src+2*stride))[0]=
0da71265
MN
1939 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1940}
1941
1942
1943#define LOAD_TOP_RIGHT_EDGE\
1944 const int t4= topright[0];\
1945 const int t5= topright[1];\
1946 const int t6= topright[2];\
1947 const int t7= topright[3];\
1948
1949#define LOAD_LEFT_EDGE\
1950 const int l0= src[-1+0*stride];\
1951 const int l1= src[-1+1*stride];\
1952 const int l2= src[-1+2*stride];\
1953 const int l3= src[-1+3*stride];\
1954
1955#define LOAD_TOP_EDGE\
1956 const int t0= src[ 0-1*stride];\
1957 const int t1= src[ 1-1*stride];\
1958 const int t2= src[ 2-1*stride];\
1959 const int t3= src[ 3-1*stride];\
1960
1961static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1962 const int lt= src[-1-1*stride];
1963 LOAD_TOP_EDGE
1964 LOAD_LEFT_EDGE
1965
115329f1 1966 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
0da71265 1967 src[0+2*stride]=
115329f1 1968 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
0da71265
MN
1969 src[0+1*stride]=
1970 src[1+2*stride]=
115329f1 1971 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
0da71265
MN
1972 src[0+0*stride]=
1973 src[1+1*stride]=
1974 src[2+2*stride]=
115329f1 1975 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
0da71265
MN
1976 src[1+0*stride]=
1977 src[2+1*stride]=
1978 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1979 src[2+0*stride]=
1980 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1981 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
4cfbf61b 1982}
0da71265
MN
1983
1984static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1985 LOAD_TOP_EDGE
1986 LOAD_TOP_RIGHT_EDGE
1987// LOAD_LEFT_EDGE
0da71265
MN
1988
1989 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1990 src[1+0*stride]=
1991 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1992 src[2+0*stride]=
1993 src[1+1*stride]=
1994 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1995 src[3+0*stride]=
1996 src[2+1*stride]=
1997 src[1+2*stride]=
1998 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1999 src[3+1*stride]=
2000 src[2+2*stride]=
2001 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
2002 src[3+2*stride]=
2003 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
2004 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
4cfbf61b 2005}
0da71265
MN
2006
2007static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
2008 const int lt= src[-1-1*stride];
115329f1
DB
2009 LOAD_TOP_EDGE
2010 LOAD_LEFT_EDGE
0da71265
MN
2011 const __attribute__((unused)) int unu= l3;
2012
2013 src[0+0*stride]=
2014 src[1+2*stride]=(lt + t0 + 1)>>1;
2015 src[1+0*stride]=
2016 src[2+2*stride]=(t0 + t1 + 1)>>1;
2017 src[2+0*stride]=
2018 src[3+2*stride]=(t1 + t2 + 1)>>1;
2019 src[3+0*stride]=(t2 + t3 + 1)>>1;
2020 src[0+1*stride]=
2021 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2022 src[1+1*stride]=
2023 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2024 src[2+1*stride]=
2025 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2026 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2027 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2028 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
4cfbf61b 2029}
0da71265
MN
2030
2031static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2032 LOAD_TOP_EDGE
2033 LOAD_TOP_RIGHT_EDGE
0da71265
MN
2034 const __attribute__((unused)) int unu= t7;
2035
2036 src[0+0*stride]=(t0 + t1 + 1)>>1;
2037 src[1+0*stride]=
2038 src[0+2*stride]=(t1 + t2 + 1)>>1;
2039 src[2+0*stride]=
2040 src[1+2*stride]=(t2 + t3 + 1)>>1;
2041 src[3+0*stride]=
2042 src[2+2*stride]=(t3 + t4+ 1)>>1;
2043 src[3+2*stride]=(t4 + t5+ 1)>>1;
2044 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2045 src[1+1*stride]=
2046 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2047 src[2+1*stride]=
2048 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2049 src[3+1*stride]=
2050 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2051 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
4cfbf61b 2052}
0da71265
MN
2053
2054static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
115329f1 2055 LOAD_LEFT_EDGE
0da71265
MN
2056
2057 src[0+0*stride]=(l0 + l1 + 1)>>1;
2058 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2059 src[2+0*stride]=
2060 src[0+1*stride]=(l1 + l2 + 1)>>1;
2061 src[3+0*stride]=
2062 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2063 src[2+1*stride]=
2064 src[0+2*stride]=(l2 + l3 + 1)>>1;
2065 src[3+1*stride]=
2066 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2067 src[3+2*stride]=
2068 src[1+3*stride]=
2069 src[0+3*stride]=
2070 src[2+2*stride]=
2071 src[2+3*stride]=
2072 src[3+3*stride]=l3;
4cfbf61b 2073}
115329f1 2074
0da71265
MN
2075static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2076 const int lt= src[-1-1*stride];
115329f1
DB
2077 LOAD_TOP_EDGE
2078 LOAD_LEFT_EDGE
0da71265
MN
2079 const __attribute__((unused)) int unu= t3;
2080
2081 src[0+0*stride]=
2082 src[2+1*stride]=(lt + l0 + 1)>>1;
2083 src[1+0*stride]=
2084 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2085 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2086 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2087 src[0+1*stride]=
2088 src[2+2*stride]=(l0 + l1 + 1)>>1;
2089 src[1+1*stride]=
2090 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2091 src[0+2*stride]=
2092 src[2+3*stride]=(l1 + l2+ 1)>>1;
2093 src[1+2*stride]=
2094 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2095 src[0+3*stride]=(l2 + l3 + 1)>>1;
2096 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
4cfbf61b 2097}
0da71265
MN
2098
2099static void pred16x16_vertical_c(uint8_t *src, int stride){
2100 int i;
2101 const uint32_t a= ((uint32_t*)(src-stride))[0];
2102 const uint32_t b= ((uint32_t*)(src-stride))[1];
2103 const uint32_t c= ((uint32_t*)(src-stride))[2];
2104 const uint32_t d= ((uint32_t*)(src-stride))[3];
115329f1 2105
0da71265
MN
2106 for(i=0; i<16; i++){
2107 ((uint32_t*)(src+i*stride))[0]= a;
2108 ((uint32_t*)(src+i*stride))[1]= b;
2109 ((uint32_t*)(src+i*stride))[2]= c;
2110 ((uint32_t*)(src+i*stride))[3]= d;
2111 }
2112}
2113
2114static void pred16x16_horizontal_c(uint8_t *src, int stride){
2115 int i;
2116
2117 for(i=0; i<16; i++){
2118 ((uint32_t*)(src+i*stride))[0]=
2119 ((uint32_t*)(src+i*stride))[1]=
2120 ((uint32_t*)(src+i*stride))[2]=
2121 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2122 }
2123}
2124
2125static void pred16x16_dc_c(uint8_t *src, int stride){
2126 int i, dc=0;
2127
2128 for(i=0;i<16; i++){
2129 dc+= src[-1+i*stride];
2130 }
115329f1 2131
0da71265
MN
2132 for(i=0;i<16; i++){
2133 dc+= src[i-stride];
2134 }
2135
2136 dc= 0x01010101*((dc + 16)>>5);
2137
2138 for(i=0; i<16; i++){
2139 ((uint32_t*)(src+i*stride))[0]=
2140 ((uint32_t*)(src+i*stride))[1]=
2141 ((uint32_t*)(src+i*stride))[2]=
2142 ((uint32_t*)(src+i*stride))[3]= dc;
2143 }
2144}
2145
2146static void pred16x16_left_dc_c(uint8_t *src, int stride){
2147 int i, dc=0;
2148
2149 for(i=0;i<16; i++){
2150 dc+= src[-1+i*stride];
2151 }
115329f1 2152
0da71265
MN
2153 dc= 0x01010101*((dc + 8)>>4);
2154
2155 for(i=0; i<16; i++){
2156 ((uint32_t*)(src+i*stride))[0]=
2157 ((uint32_t*)(src+i*stride))[1]=
2158 ((uint32_t*)(src+i*stride))[2]=
2159 ((uint32_t*)(src+i*stride))[3]= dc;
2160 }
2161}
2162
2163static void pred16x16_top_dc_c(uint8_t *src, int stride){
2164 int i, dc=0;
2165
2166 for(i=0;i<16; i++){
2167 dc+= src[i-stride];
2168 }
2169 dc= 0x01010101*((dc + 8)>>4);
2170
2171 for(i=0; i<16; i++){
2172 ((uint32_t*)(src+i*stride))[0]=
2173 ((uint32_t*)(src+i*stride))[1]=
2174 ((uint32_t*)(src+i*stride))[2]=
2175 ((uint32_t*)(src+i*stride))[3]= dc;
2176 }
2177}
2178
2179static void pred16x16_128_dc_c(uint8_t *src, int stride){
2180 int i;
2181
2182 for(i=0; i<16; i++){
2183 ((uint32_t*)(src+i*stride))[0]=
2184 ((uint32_t*)(src+i*stride))[1]=
2185 ((uint32_t*)(src+i*stride))[2]=
2186 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2187 }
2188}
2189
8b82a956 2190static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
30f73fc7
MN
2191 int i, j, k;
2192 int a;
2193 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2194 const uint8_t * const src0 = src+7-stride;
2195 const uint8_t *src1 = src+8*stride-1;
2196 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2197 int H = src0[1] - src0[-1];
2198 int V = src1[0] - src2[ 0];
2199 for(k=2; k<=8; ++k) {
2200 src1 += stride; src2 -= stride;
2201 H += k*(src0[k] - src0[-k]);
2202 V += k*(src1[0] - src2[ 0]);
2203 }
8b82a956
MN
2204 if(svq3){
2205 H = ( 5*(H/4) ) / 16;
2206 V = ( 5*(V/4) ) / 16;
2e26c8d2
MM
2207
2208 /* required for 100% accuracy */
2209 i = H; H = V; V = i;
8b82a956
MN
2210 }else{
2211 H = ( 5*H+32 ) >> 6;
2212 V = ( 5*V+32 ) >> 6;
2213 }
30f73fc7
MN
2214
2215 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2216 for(j=16; j>0; --j) {
2217 int b = a;
2218 a += V;
2219 for(i=-16; i<0; i+=4) {
2220 src[16+i] = cm[ (b ) >> 5 ];
2221 src[17+i] = cm[ (b+ H) >> 5 ];
2222 src[18+i] = cm[ (b+2*H) >> 5 ];
2223 src[19+i] = cm[ (b+3*H) >> 5 ];
2224 b += 4*H;
0da71265 2225 }
30f73fc7
MN
2226 src += stride;
2227 }
0da71265
MN
2228}
2229
8b82a956
MN
2230static void pred16x16_plane_c(uint8_t *src, int stride){
2231 pred16x16_plane_compat_c(src, stride, 0);
2232}
2233
0da71265
MN
2234static void pred8x8_vertical_c(uint8_t *src, int stride){
2235 int i;
2236 const uint32_t a= ((uint32_t*)(src-stride))[0];
2237 const uint32_t b= ((uint32_t*)(src-stride))[1];
115329f1 2238
0da71265
MN
2239 for(i=0; i<8; i++){
2240 ((uint32_t*)(src+i*stride))[0]= a;
2241 ((uint32_t*)(src+i*stride))[1]= b;
2242 }
2243}
2244
2245static void pred8x8_horizontal_c(uint8_t *src, int stride){
2246 int i;
2247
2248 for(i=0; i<8; i++){
2249 ((uint32_t*)(src+i*stride))[0]=
2250 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2251 }
2252}
2253
2254static void pred8x8_128_dc_c(uint8_t *src, int stride){
2255 int i;
2256
43efd19a 2257 for(i=0; i<8; i++){
115329f1 2258 ((uint32_t*)(src+i*stride))[0]=
0da71265
MN
2259 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2260 }
2261}
2262
2263static void pred8x8_left_dc_c(uint8_t *src, int stride){
2264 int i;
2265 int dc0, dc2;
2266
2267 dc0=dc2=0;
2268 for(i=0;i<4; i++){
2269 dc0+= src[-1+i*stride];
2270 dc2+= src[-1+(i+4)*stride];
2271 }
2272 dc0= 0x01010101*((dc0 + 2)>>2);
2273 dc2= 0x01010101*((dc2 + 2)>>2);
2274
2275 for(i=0; i<4; i++){
2276 ((uint32_t*)(src+i*stride))[0]=
2277 ((uint32_t*)(src+i*stride))[1]= dc0;
2278 }
2279 for(i=4; i<8; i++){
2280 ((uint32_t*)(src+i*stride))[0]=
2281 ((uint32_t*)(src+i*stride))[1]= dc2;
2282 }
2283}
2284
2285static void pred8x8_top_dc_c(uint8_t *src, int stride){
2286 int i;
2287 int dc0, dc1;
2288
2289 dc0=dc1=0;
2290 for(i=0;i<4; i++){
2291 dc0+= src[i-stride];
2292 dc1+= src[4+i-stride];
2293 }
2294 dc0= 0x01010101*((dc0 + 2)>>2);
2295 dc1= 0x01010101*((dc1 + 2)>>2);
2296
2297 for(i=0; i<4; i++){
2298 ((uint32_t*)(src+i*stride))[0]= dc0;
2299 ((uint32_t*)(src+i*stride))[1]= dc1;
2300 }
2301 for(i=4; i<8; i++){
2302 ((uint32_t*)(src+i*stride))[0]= dc0;
2303 ((uint32_t*)(src+i*stride))[1]= dc1;
2304 }
2305}
2306
2307
2308static void pred8x8_dc_c(uint8_t *src, int stride){
2309 int i;
2310 int dc0, dc1, dc2, dc3;
2311
2312 dc0=dc1=dc2=0;
2313 for(i=0;i<4; i++){
2314 dc0+= src[-1+i*stride] + src[i-stride];
2315 dc1+= src[4+i-stride];
2316 dc2+= src[-1+(i+4)*stride];
2317 }
2318 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2319 dc0= 0x01010101*((dc0 + 4)>>3);
2320 dc1= 0x01010101*((dc1 + 2)>>2);
2321 dc2= 0x01010101*((dc2 + 2)>>2);
2322
2323 for(i=0; i<4; i++){
2324 ((uint32_t*)(src+i*stride))[0]= dc0;
2325 ((uint32_t*)(src+i*stride))[1]= dc1;
2326 }
2327 for(i=4; i<8; i++){
2328 ((uint32_t*)(src+i*stride))[0]= dc2;
2329 ((uint32_t*)(src+i*stride))[1]= dc3;
2330 }
2331}
2332
2333static void pred8x8_plane_c(uint8_t *src, int stride){
30f73fc7
MN
2334 int j, k;
2335 int a;
2336 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2337 const uint8_t * const src0 = src+3-stride;
2338 const uint8_t *src1 = src+4*stride-1;
2339 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2340 int H = src0[1] - src0[-1];
2341 int V = src1[0] - src2[ 0];
2342 for(k=2; k<=4; ++k) {
2343 src1 += stride; src2 -= stride;
2344 H += k*(src0[k] - src0[-k]);
2345 V += k*(src1[0] - src2[ 0]);
2346 }
2347 H = ( 17*H+16 ) >> 5;
2348 V = ( 17*V+16 ) >> 5;
2349
2350 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2351 for(j=8; j>0; --j) {
2352 int b = a;
2353 a += V;
2354 src[0] = cm[ (b ) >> 5 ];
2355 src[1] = cm[ (b+ H) >> 5 ];
2356 src[2] = cm[ (b+2*H) >> 5 ];
2357 src[3] = cm[ (b+3*H) >> 5 ];
2358 src[4] = cm[ (b+4*H) >> 5 ];
2359 src[5] = cm[ (b+5*H) >> 5 ];
2360 src[6] = cm[ (b+6*H) >> 5 ];
2361 src[7] = cm[ (b+7*H) >> 5 ];
2362 src += stride;
2363 }
0da71265
MN
2364}
2365
43efd19a
LM
2366#define SRC(x,y) src[(x)+(y)*stride]
2367#define PL(y) \
2368 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2369#define PREDICT_8x8_LOAD_LEFT \
2370 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2371 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2372 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
bf4e3bd2 2373 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
43efd19a
LM
2374
2375#define PT(x) \
2376 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2377#define PREDICT_8x8_LOAD_TOP \
2378 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2379 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2380 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
bf4e3bd2 2381 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
fb3ca6ba 2382 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
43efd19a
LM
2383
2384#define PTR(x) \
2385 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2386#define PREDICT_8x8_LOAD_TOPRIGHT \
2387 int t8, t9, t10, t11, t12, t13, t14, t15; \
2388 if(has_topright) { \
2389 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2390 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2391 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2392
2393#define PREDICT_8x8_LOAD_TOPLEFT \
fb3ca6ba 2394 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
43efd19a
LM
2395
2396#define PREDICT_8x8_DC(v) \
2397 int y; \
2398 for( y = 0; y < 8; y++ ) { \
2399 ((uint32_t*)src)[0] = \
2400 ((uint32_t*)src)[1] = v; \
2401 src += stride; \
2402 }
2403
2404static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2405{
2406 PREDICT_8x8_DC(0x80808080);
2407}
2408static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2409{
2410 PREDICT_8x8_LOAD_LEFT;
2411 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2412 PREDICT_8x8_DC(dc);
2413}
2414static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2415{
2416 PREDICT_8x8_LOAD_TOP;
2417 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2418 PREDICT_8x8_DC(dc);
2419}
2420static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2421{
2422 PREDICT_8x8_LOAD_LEFT;
2423 PREDICT_8x8_LOAD_TOP;
2424 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2425 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2426 PREDICT_8x8_DC(dc);
2427}
2428static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2429{
2430 PREDICT_8x8_LOAD_LEFT;
2431#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2432 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2433 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2434#undef ROW
2435}
2436static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2437{
2438 int y;
2439 PREDICT_8x8_LOAD_TOP;
2440 src[0] = t0;
2441 src[1] = t1;
2442 src[2] = t2;
2443 src[3] = t3;
2444 src[4] = t4;
2445 src[5] = t5;
2446 src[6] = t6;
2447 src[7] = t7;
2448 for( y = 1; y < 8; y++ )
2449 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2450}
2451static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2452{
2453 PREDICT_8x8_LOAD_TOP;
2454 PREDICT_8x8_LOAD_TOPRIGHT;
2455 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2456 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2457 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2458 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2459 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2460 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2461 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;
2462 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;
2463 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;
2464 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2465 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2466 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2467 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2468 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2469 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2470}
2471static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2472{
2473 PREDICT_8x8_LOAD_TOP;
2474 PREDICT_8x8_LOAD_LEFT;
2475 PREDICT_8x8_LOAD_TOPLEFT;
2476 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2477 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2478 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2479 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2480 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2481 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2482 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;
2483 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;
2484 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;
2485 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2486 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2487 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2488 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2489 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2490 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
115329f1 2491
43efd19a
LM
2492}
2493static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2494{
2495 PREDICT_8x8_LOAD_TOP;
2496 PREDICT_8x8_LOAD_LEFT;
2497 PREDICT_8x8_LOAD_TOPLEFT;
2498 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2499 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2500 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2501 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2502 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2503 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2504 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2505 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2506 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2507 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2508 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2509 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2510 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2511 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2512 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2513 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2514 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2515 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2516 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2517 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2518 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2519 SRC(7,0)= (t6 + t7 + 1) >> 1;
2520}
2521static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2522{
2523 PREDICT_8x8_LOAD_TOP;
2524 PREDICT_8x8_LOAD_LEFT;
2525 PREDICT_8x8_LOAD_TOPLEFT;
2526 SRC(0,7)= (l6 + l7 + 1) >> 1;
2527 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2528 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2529 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2530 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2531 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2532 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2533 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2534 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2535 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2536 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2537 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2538 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2539 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2540 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2541 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2542 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2543 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2544 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2545 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2546 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2547 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2548}
2549static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2550{
2551 PREDICT_8x8_LOAD_TOP;
2552 PREDICT_8x8_LOAD_TOPRIGHT;
2553 SRC(0,0)= (t0 + t1 + 1) >> 1;
2554 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2555 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2556 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2557 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2558 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2559 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2560 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2561 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2562 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2563 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2564 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2565 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2566 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2567 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2568 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2569 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2570 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2571 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2572 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2573 SRC(7,6)= (t10 + t11 + 1) >> 1;
2574 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2575}
2576static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2577{
2578 PREDICT_8x8_LOAD_LEFT;
2579 SRC(0,0)= (l0 + l1 + 1) >> 1;
2580 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2581 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2582 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2583 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2584 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2585 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2586 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2587 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2588 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2589 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2590 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2591 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2592 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2593 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2594 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2595 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2596 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2597}
2598#undef PREDICT_8x8_LOAD_LEFT
2599#undef PREDICT_8x8_LOAD_TOP
2600#undef PREDICT_8x8_LOAD_TOPLEFT
2601#undef PREDICT_8x8_LOAD_TOPRIGHT
2602#undef PREDICT_8x8_DC
2603#undef PTR
2604#undef PT
2605#undef PL
2606#undef SRC
2607
0da71265
MN
2608static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2609 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2610 int src_x_offset, int src_y_offset,
2611 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2612 MpegEncContext * const s = &h->s;
2613 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2614 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2615 const int luma_xy= (mx&3) + ((my&3)<<2);
2616 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2617 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2618 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2619 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2620 int extra_height= extra_width;
2621 int emu=0;
2622 const int full_mx= mx>>2;
2623 const int full_my= my>>2;
fbd312fd
LM
2624 const int pic_width = 16*s->mb_width;
2625 const int pic_height = 16*s->mb_height;
115329f1 2626
171c4076
LM
2627 if(!pic->data[0])
2628 return;
115329f1 2629
0da71265
MN
2630 if(mx&7) extra_width -= 3;
2631 if(my&7) extra_height -= 3;
115329f1
DB
2632
2633 if( full_mx < 0-extra_width
2634 || full_my < 0-extra_height
2635 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd
LM
2636 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2637 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
0da71265
MN
2638 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2639 emu=1;
2640 }
115329f1 2641
0da71265
MN
2642 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2643 if(!square){
2644 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2645 }
115329f1 2646
0da71265 2647 if(s->flags&CODEC_FLAG_GRAY) return;
115329f1 2648
0da71265 2649 if(emu){
fbd312fd 2650 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
2651 src_cb= s->edge_emu_buffer;
2652 }
2653 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2654
2655 if(emu){
fbd312fd 2656 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
0da71265
MN
2657 src_cr= s->edge_emu_buffer;
2658 }
2659 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2660}
2661
9f2d1b4f 2662static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
2663 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2664 int x_offset, int y_offset,
2665 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2666 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2667 int list0, int list1){
2668 MpegEncContext * const s = &h->s;
2669 qpel_mc_func *qpix_op= qpix_put;
2670 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 2671
0da71265
MN
2672 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2673 dest_cb += x_offset + y_offset*s->uvlinesize;
2674 dest_cr += x_offset + y_offset*s->uvlinesize;
2675 x_offset += 8*s->mb_x;
2676 y_offset += 8*s->mb_y;
115329f1 2677
0da71265 2678 if(list0){
1924f3ce 2679 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
2680 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2681 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2682 qpix_op, chroma_op);
2683
2684 qpix_op= qpix_avg;
2685 chroma_op= chroma_avg;
2686 }
2687
2688 if(list1){
1924f3ce 2689 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
2690 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2691 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2692 qpix_op, chroma_op);
2693 }
2694}
2695
9f2d1b4f
LM
2696static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2697 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2698 int x_offset, int y_offset,
2699 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2700 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2701 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2702 int list0, int list1){
2703 MpegEncContext * const s = &h->s;
2704
2705 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2706 dest_cb += x_offset + y_offset*s->uvlinesize;
2707 dest_cr += x_offset + y_offset*s->uvlinesize;
2708 x_offset += 8*s->mb_x;
2709 y_offset += 8*s->mb_y;
115329f1 2710
9f2d1b4f
LM
2711 if(list0 && list1){
2712 /* don't optimize for luma-only case, since B-frames usually
2713 * use implicit weights => chroma too. */
2714 uint8_t *tmp_cb = s->obmc_scratchpad;
2715 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2716 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2717 int refn0 = h->ref_cache[0][ scan8[n] ];
2718 int refn1 = h->ref_cache[1][ scan8[n] ];
2719
2720 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2721 dest_y, dest_cb, dest_cr,
2722 x_offset, y_offset, qpix_put, chroma_put);
2723 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2724 tmp_y, tmp_cb, tmp_cr,
2725 x_offset, y_offset, qpix_put, chroma_put);
2726
2727 if(h->use_weight == 2){
2728 int weight0 = h->implicit_weight[refn0][refn1];
2729 int weight1 = 64 - weight0;
e8b56208
LM
2730 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0);
2731 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2732 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f
LM
2733 }else{
2734 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
115329f1 2735 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
e8b56208 2736 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
9f2d1b4f 2737 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2738 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
e8b56208 2739 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
9f2d1b4f 2740 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2741 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
e8b56208 2742 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
9f2d1b4f
LM
2743 }
2744 }else{
2745 int list = list1 ? 1 : 0;
2746 int refn = h->ref_cache[list][ scan8[n] ];
2747 Picture *ref= &h->ref_list[list][refn];
2748 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2749 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2750 qpix_put, chroma_put);
2751
2752 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2753 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2754 if(h->use_weight_chroma){
2755 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2756 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2757 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2758 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2759 }
2760 }
2761}
2762
2763static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2764 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2765 int x_offset, int y_offset,
2766 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2767 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 2768 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
9f2d1b4f
LM
2769 int list0, int list1){
2770 if((h->use_weight==2 && list0 && list1
2771 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2772 || h->use_weight==1)
2773 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2774 x_offset, y_offset, qpix_put, chroma_put,
2775 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2776 else
2777 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2778 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2779}
2780
513fbd8e
LM
2781static inline void prefetch_motion(H264Context *h, int list){
2782 /* fetch pixels for estimated mv 4 macroblocks ahead
2783 * optimized for 64byte cache lines */
2784 MpegEncContext * const s = &h->s;
2785 const int refn = h->ref_cache[list][scan8[0]];
2786 if(refn >= 0){
2787 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2788 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2789 uint8_t **src= h->ref_list[list][refn].data;
2790 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
2791 s->dsp.prefetch(src[0]+off, s->linesize, 4);
2792 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2793 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2794 }
2795}
2796
0da71265
MN
2797static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2798 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
2799 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2800 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 2801 MpegEncContext * const s = &h->s;
7bc9090a 2802 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 2803 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 2804
0da71265 2805 assert(IS_INTER(mb_type));
115329f1 2806
513fbd8e
LM
2807 prefetch_motion(h, 0);
2808
0da71265
MN
2809 if(IS_16X16(mb_type)){
2810 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2811 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 2812 &weight_op[0], &weight_avg[0],
0da71265
MN
2813 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2814 }else if(IS_16X8(mb_type)){
2815 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2816 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2817 &weight_op[1], &weight_avg[1],
0da71265
MN
2818 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2819 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2820 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2821 &weight_op[1], &weight_avg[1],
0da71265
MN
2822 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2823 }else if(IS_8X16(mb_type)){
2824 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2825 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2826 &weight_op[2], &weight_avg[2],
0da71265
MN
2827 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2828 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2829 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2830 &weight_op[2], &weight_avg[2],
0da71265
MN
2831 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2832 }else{
2833 int i;
115329f1 2834
0da71265
MN
2835 assert(IS_8X8(mb_type));
2836
2837 for(i=0; i<4; i++){
2838 const int sub_mb_type= h->sub_mb_type[i];
2839 const int n= 4*i;
2840 int x_offset= (i&1)<<2;
2841 int y_offset= (i&2)<<1;
2842
2843 if(IS_SUB_8X8(sub_mb_type)){
2844 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2845 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2846 &weight_op[3], &weight_avg[3],
0da71265
MN
2847 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2848 }else if(IS_SUB_8X4(sub_mb_type)){
2849 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2850 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2851 &weight_op[4], &weight_avg[4],
0da71265
MN
2852 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2853 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2854 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2855 &weight_op[4], &weight_avg[4],
0da71265
MN
2856 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2857 }else if(IS_SUB_4X8(sub_mb_type)){
2858 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2859 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2860 &weight_op[5], &weight_avg[5],
0da71265
MN
2861 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2862 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2863 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2864 &weight_op[5], &weight_avg[5],
0da71265
MN
2865 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2866 }else{
2867 int j;
2868 assert(IS_SUB_4X4(sub_mb_type));
2869 for(j=0; j<4; j++){
2870 int sub_x_offset= x_offset + 2*(j&1);
2871 int sub_y_offset= y_offset + (j&2);
2872 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2873 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2874 &weight_op[6], &weight_avg[6],
0da71265
MN
2875 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2876 }
2877 }
2878 }
2879 }
513fbd8e
LM
2880
2881 prefetch_motion(h, 1);
0da71265
MN
2882}
2883
2884static void decode_init_vlc(H264Context *h){
2885 static int done = 0;
2886
2887 if (!done) {
2888 int i;
2889 done = 1;
2890
115329f1 2891 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
0da71265 2892 &chroma_dc_coeff_token_len [0], 1, 1,
073c2593 2893 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
0da71265
MN
2894
2895 for(i=0; i<4; i++){
115329f1 2896 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
0da71265 2897 &coeff_token_len [i][0], 1, 1,
073c2593 2898 &coeff_token_bits[i][0], 1, 1, 1);
0da71265
MN
2899 }
2900
2901 for(i=0; i<3; i++){
2902 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2903 &chroma_dc_total_zeros_len [i][0], 1, 1,
073c2593 2904 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2905 }
2906 for(i=0; i<15; i++){
115329f1 2907 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
0da71265 2908 &total_zeros_len [i][0], 1, 1,
073c2593 2909 &total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2910 }
2911
2912 for(i=0; i<6; i++){
115329f1 2913 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
0da71265 2914 &run_len [i][0], 1, 1,
073c2593 2915 &run_bits[i][0], 1, 1, 1);
0da71265 2916 }
115329f1 2917 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
0da71265 2918 &run_len [6][0], 1, 1,
073c2593 2919 &run_bits[6][0], 1, 1, 1);
0da71265
MN
2920 }
2921}
2922
2923/**
2924 * Sets the intra prediction function pointers.
2925 */
2926static void init_pred_ptrs(H264Context *h){
2927// MpegEncContext * const s = &h->s;
2928
2929 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2930 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2931 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2932 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2933 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2934 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2935 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2936 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2937 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2938 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2939 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2940 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2941
43efd19a
LM
2942 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
2943 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
2944 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
2945 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2946 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2947 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
2948 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
2949 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
2950 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
2951 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
2952 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
2953 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
2954
0da71265
MN
2955 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2956 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2957 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2958 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2959 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2960 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2961 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2962
2963 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2964 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2965 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2966 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2967 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2968 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2969 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2970}
2971
0da71265 2972static void free_tables(H264Context *h){
0da71265 2973 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
2974 av_freep(&h->chroma_pred_mode_table);
2975 av_freep(&h->cbp_table);
9e528114
LA
2976 av_freep(&h->mvd_table[0]);
2977 av_freep(&h->mvd_table[1]);
5ad984c9 2978 av_freep(&h->direct_table);
0da71265
MN
2979 av_freep(&h->non_zero_count);
2980 av_freep(&h->slice_table_base);
6ba71fc4
LLL
2981 av_freep(&h->top_borders[1]);
2982 av_freep(&h->top_borders[0]);
0da71265 2983 h->slice_table= NULL;
e5017ab8 2984
0da71265
MN
2985 av_freep(&h->mb2b_xy);
2986 av_freep(&h->mb2b8_xy);
9f2d1b4f
LM
2987
2988 av_freep(&h->s.obmc_scratchpad);
0da71265
MN
2989}
2990
239ea04c
LM
2991static void init_dequant8_coeff_table(H264Context *h){
2992 int i,q,x;
548a1c8a 2993 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
239ea04c
LM
2994 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2995 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2996
2997 for(i=0; i<2; i++ ){
2998 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2999 h->dequant8_coeff[1] = h->dequant8_buffer[0];
3000 break;
3001 }
3002
3003 for(q=0; q<52; q++){
3004 int shift = div6[q];
3005 int idx = rem6[q];
3006 for(x=0; x<64; x++)
548a1c8a
LM
3007 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
3008 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
3009 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
3010 }
3011 }
3012}
3013
3014static void init_dequant4_coeff_table(H264Context *h){
3015 int i,j,q,x;
ab2e3e2c 3016 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
239ea04c
LM
3017 for(i=0; i<6; i++ ){
3018 h->dequant4_coeff[i] = h->dequant4_buffer[i];
3019 for(j=0; j<i; j++){
3020 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
3021 h->dequant4_coeff[i] = h->dequant4_buffer[j];
3022 break;
3023 }
3024 }
3025 if(j<i)
3026 continue;
3027
3028 for(q=0; q<52; q++){
3029 int shift = div6[q] + 2;
3030 int idx = rem6[q];
3031 for(x=0; x<16; x++)
ab2e3e2c
LM
3032 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3033 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
3034 h->pps.scaling_matrix4[i][x]) << shift;
3035 }
3036 }
3037}
3038
3039static void init_dequant_tables(H264Context *h){
3040 int i,x;
3041 init_dequant4_coeff_table(h);
3042 if(h->pps.transform_8x8_mode)
3043 init_dequant8_coeff_table(h);
3044 if(h->sps.transform_bypass){
3045 for(i=0; i<6; i++)
3046 for(x=0; x<16; x++)
3047 h->dequant4_coeff[i][0][x] = 1<<6;
3048 if(h->pps.transform_8x8_mode)
3049 for(i=0; i<2; i++)
3050 for(x=0; x<64; x++)
3051 h->dequant8_coeff[i][0][x] = 1<<6;
3052 }
3053}
3054
3055
0da71265
MN
3056/**
3057 * allocates tables.
3b66c4c5 3058 * needs width/height
0da71265
MN
3059 */
3060static int alloc_tables(H264Context *h){
3061 MpegEncContext * const s = &h->s;
7bc9090a 3062 const int big_mb_num= s->mb_stride * (s->mb_height+1);
239ea04c 3063 int x,y;
0da71265
MN
3064
3065 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 3066
53c05b1e 3067 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
0da71265 3068 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
6ba71fc4
LLL
3069 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
3070 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
5d0e4cb8 3071 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 3072
e5017ab8
LA
3073 if( h->pps.cabac ) {
3074 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
9e528114
LA
3075 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3076 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
5ad984c9 3077 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8
LA
3078 }
3079
0da71265 3080 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
7bc9090a 3081 h->slice_table= h->slice_table_base + s->mb_stride + 1;
0da71265 3082
a55f20bd
LM
3083 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
3084 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
0da71265
MN
3085 for(y=0; y<s->mb_height; y++){
3086 for(x=0; x<s->mb_width; x++){
7bc9090a 3087 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
3088 const int b_xy = 4*x + 4*y*h->b_stride;
3089 const int b8_xy= 2*x + 2*y*h->b8_stride;
115329f1 3090
0da71265
MN
3091 h->mb2b_xy [mb_xy]= b_xy;
3092 h->mb2b8_xy[mb_xy]= b8_xy;
3093 }
3094 }
9f2d1b4f 3095
9c6221ae
GV
3096 s->obmc_scratchpad = NULL;
3097
56edbd81
LM
3098 if(!h->dequant4_coeff[0])
3099 init_dequant_tables(h);
3100
0da71265
MN
3101 return 0;
3102fail:
3103 free_tables(h);
3104 return -1;
3105}
3106
3107static void common_init(H264Context *h){
3108 MpegEncContext * const s = &h->s;
0da71265
MN
3109
3110 s->width = s->avctx->width;
3111 s->height = s->avctx->height;
3112 s->codec_id= s->avctx->codec->id;
115329f1 3113
0da71265
MN
3114 init_pred_ptrs(h);
3115
239ea04c 3116 h->dequant_coeff_pps= -1;
9a41c2c7 3117 s->unrestricted_mv=1;
0da71265 3118 s->decode=1; //FIXME
56edbd81
LM
3119
3120 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3121 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
3122}
3123
3124static int decode_init(AVCodecContext *avctx){
3125 H264Context *h= avctx->priv_data;
3126 MpegEncContext * const s = &h->s;
3127
3edcacde 3128 MPV_decode_defaults(s);
115329f1 3129
0da71265
MN
3130 s->avctx = avctx;
3131 common_init(h);
3132
3133 s->out_format = FMT_H264;
3134 s->workaround_bugs= avctx->workaround_bugs;
3135
3136 // set defaults
0da71265
MN
3137// s->decode_mb= ff_h263_decode_mb;
3138 s->low_delay= 1;
3139 avctx->pix_fmt= PIX_FMT_YUV420P;
3140
3141 decode_init_vlc(h);
115329f1 3142
26165f99
MR
3143 if(avctx->extradata_size > 0 && avctx->extradata &&
3144 *(char *)avctx->extradata == 1){
4770b1b4
RT
3145 h->is_avc = 1;
3146 h->got_avcC = 0;
26165f99
MR
3147 } else {
3148 h->is_avc = 0;
4770b1b4
RT
3149 }
3150
0da71265
MN
3151 return 0;
3152}
3153
af8aa846 3154static int frame_start(H264Context *h){
0da71265
MN
3155 MpegEncContext * const s = &h->s;
3156 int i;
3157
af8aa846
MN
3158 if(MPV_frame_start(s, s->avctx) < 0)
3159 return -1;
0da71265 3160 ff_er_frame_start(s);
0da71265
MN
3161
3162 assert(s->linesize && s->uvlinesize);
3163
3164 for(i=0; i<16; i++){
3165 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 3166 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3167 }
3168 for(i=0; i<4; i++){
3169 h->block_offset[16+i]=
3170 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
3171 h->block_offset[24+16+i]=
3172 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3173 }
3174
934b0821
LM
3175 /* can't be in alloc_tables because linesize isn't known there.
3176 * FIXME: redo bipred weight to not require extra buffer? */
3177 if(!s->obmc_scratchpad)
3178 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3179
0da71265 3180// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
af8aa846 3181 return 0;
0da71265
MN
3182}
3183
53c05b1e
MN
3184static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3185 MpegEncContext * const s = &h->s;
3186 int i;
115329f1 3187
53c05b1e
MN
3188 src_y -= linesize;
3189 src_cb -= uvlinesize;
3190 src_cr -= uvlinesize;
3191
3b66c4c5 3192 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 3193 // and the line above the bottom macroblock
6ba71fc4 3194 h->left_border[0]= h->top_borders[0][s->mb_x][15];
53c05b1e
MN
3195 for(i=1; i<17; i++){
3196 h->left_border[i]= src_y[15+i* linesize];
3197 }
115329f1 3198
6ba71fc4
LLL
3199 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3200 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
53c05b1e
MN
3201
3202 if(!(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
3203 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3204 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
53c05b1e
MN
3205 for(i=1; i<9; i++){
3206 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3207 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3208 }
6ba71fc4
LLL
3209 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3210 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
53c05b1e
MN
3211 }
3212}
3213
3214static 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){
3215 MpegEncContext * const s = &h->s;
3216 int temp8, i;
3217 uint64_t temp64;
d89dc06a
LM
3218 int deblock_left = (s->mb_x > 0);
3219 int deblock_top = (s->mb_y > 0);
53c05b1e
MN
3220
3221 src_y -= linesize + 1;
3222 src_cb -= uvlinesize + 1;
3223 src_cr -= uvlinesize + 1;
3224
3225#define XCHG(a,b,t,xchg)\
3226t= a;\
3227if(xchg)\
3228 a= b;\
3229b= t;
d89dc06a
LM
3230
3231 if(deblock_left){
3232 for(i = !deblock_top; i<17; i++){
3233 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3234 }
3235 }
3236
3237 if(deblock_top){
6ba71fc4
LLL
3238 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3239 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
cad4368a 3240 if(s->mb_x+1 < s->mb_width){
43efd19a
LM
3241 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3242 }
53c05b1e 3243 }
53c05b1e
MN
3244
3245 if(!(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
3246 if(deblock_left){
3247 for(i = !deblock_top; i<9; i++){
3248 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3249 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3250 }
3251 }
3252 if(deblock_top){
6ba71fc4
LLL
3253 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3254 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3255 }
3256 }
3257}
3258
3259static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3260 MpegEncContext * const s = &h->s;
3261 int i;
115329f1 3262
6ba71fc4
LLL
3263 src_y -= 2 * linesize;
3264 src_cb -= 2 * uvlinesize;
3265 src_cr -= 2 * uvlinesize;
3266
3b66c4c5 3267 // There are two lines saved, the line above the the top macroblock of a pair,
6ba71fc4
LLL
3268 // and the line above the bottom macroblock
3269 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3270 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3271 for(i=2; i<34; i++){
3272 h->left_border[i]= src_y[15+i* linesize];
3273 }
115329f1 3274
6ba71fc4
LLL
3275 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3276 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3277 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3278 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3279
3280 if(!(s->flags&CODEC_FLAG_GRAY)){
3281 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3282 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3283 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3284 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3285 for(i=2; i<18; i++){
3286 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3287 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3288 }
3289 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3290 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3291 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3292 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3293 }
3294}
3295
3296static 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){
3297 MpegEncContext * const s = &h->s;
3298 int temp8, i;
3299 uint64_t temp64;
3300 int deblock_left = (s->mb_x > 0);
3301 int deblock_top = (s->mb_y > 0);
3302
3303 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
3304
3305 src_y -= 2 * linesize + 1;
3306 src_cb -= 2 * uvlinesize + 1;
3307 src_cr -= 2 * uvlinesize + 1;
3308
3309#define XCHG(a,b,t,xchg)\
3310t= a;\
3311if(xchg)\
3312 a= b;\
3313b= t;
3314
3315 if(deblock_left){
3316 for(i = (!deblock_top)<<1; i<34; i++){
3317 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3318 }
3319 }
3320
3321 if(deblock_top){
3322 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3323 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3324 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3325 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3326 }
3327
3328 if(!(s->flags&CODEC_FLAG_GRAY)){
3329 if(deblock_left){
3330 for(i = (!deblock_top) << 1; i<18; i++){
3331 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
3332 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3333 }
3334 }
3335 if(deblock_top){
3336 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3337 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3338 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3339 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
53c05b1e 3340 }
53c05b1e
MN
3341 }
3342}
3343
0da71265
MN
3344static void hl_decode_mb(H264Context *h){
3345 MpegEncContext * const s = &h->s;
3346 const int mb_x= s->mb_x;
3347 const int mb_y= s->mb_y;
7bc9090a 3348 const int mb_xy= mb_x + mb_y*s->mb_stride;
0da71265
MN
3349 const int mb_type= s->current_picture.mb_type[mb_xy];
3350 uint8_t *dest_y, *dest_cb, *dest_cr;
3351 int linesize, uvlinesize /*dct_offset*/;
3352 int i;
6867a90b 3353 int *block_offset = &h->block_offset[0];
6ba71fc4 3354 const unsigned int bottom = mb_y & 1;
36940eca
LM
3355 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3356 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15 3357 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
0da71265
MN
3358
3359 if(!s->decode)
3360 return;
3361
0da71265
MN
3362 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3363 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3364 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3365
3366 if (h->mb_field_decoding_flag) {
3367 linesize = s->linesize * 2;
3368 uvlinesize = s->uvlinesize * 2;
6867a90b 3369 block_offset = &h->block_offset[24];
0da71265
MN
3370 if(mb_y&1){ //FIXME move out of this func?
3371 dest_y -= s->linesize*15;
6867a90b
LLL
3372 dest_cb-= s->uvlinesize*7;
3373 dest_cr-= s->uvlinesize*7;
0da71265
MN
3374 }
3375 } else {
3376 linesize = s->linesize;
3377 uvlinesize = s->uvlinesize;
3378// dct_offset = s->linesize * 16;
3379 }
115329f1 3380
ef9d1d15
LM
3381 if(transform_bypass){
3382 idct_dc_add =
3383 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3384 }else if(IS_8x8DCT(mb_type)){
3385 idct_dc_add = s->dsp.h264_idct8_dc_add;
3386 idct_add = s->dsp.h264_idct8_add;
3387 }else{
3388 idct_dc_add = s->dsp.h264_idct_dc_add;
3389 idct_add = s->dsp.h264_idct_add;
3390 }
0da71265 3391
6fbcaaa0
LLL
3392 if (IS_INTRA_PCM(mb_type)) {
3393 unsigned int x, y;
3394
3395 // The pixels are stored in h->mb array in the same order as levels,
3396 // copy them in output in the correct order.
3397 for(i=0; i<16; i++) {
3398 for (y=0; y<4; y++) {
3399 for (x=0; x<4; x++) {
6867a90b 3400 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3401 }
3402 }
3403 }
3404 for(i=16; i<16+4; i++) {
3405 for (y=0; y<4; y++) {
3406 for (x=0; x<4; x++) {
6867a90b 3407 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3408 }
3409 }
3410 }
3411 for(i=20; i<20+4; i++) {
3412 for (y=0; y<4; y++) {
3413 for (x=0; x<4; x++) {
6867a90b 3414 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3415 }
3416 }
3417 }
e7e09b49
LLL
3418 } else {
3419 if(IS_INTRA(mb_type)){
6ba71fc4
LLL
3420 if(h->deblocking_filter) {
3421 if (h->mb_aff_frame) {
3422 if (!bottom)
3423 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3424 } else {
3425 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3426 }
3427 }
53c05b1e 3428
e7e09b49
LLL
3429 if(!(s->flags&CODEC_FLAG_GRAY)){
3430 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3431 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3432 }
0da71265 3433
e7e09b49
LLL
3434 if(IS_INTRA4x4(mb_type)){
3435 if(!s->encoding){
43efd19a
LM
3436 if(IS_8x8DCT(mb_type)){
3437 for(i=0; i<16; i+=4){
3438 uint8_t * const ptr= dest_y + block_offset[i];
3439 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3440 const int nnz = h->non_zero_count_cache[ scan8[i] ];
43efd19a
LM
3441 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3442 (h->topright_samples_available<<(i+1))&0x8000, linesize);
ef9d1d15
LM
3443 if(nnz){
3444 if(nnz == 1 && h->mb[i*16])
3445 idct_dc_add(ptr, h->mb + i*16, linesize);
3446 else
3447 idct_add(ptr, h->mb + i*16, linesize);
3448 }
43efd19a
LM
3449 }
3450 }else
e7e09b49 3451 for(i=0; i<16; i++){
6867a90b 3452 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
3453 uint8_t *topright;
3454 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3455 int nnz, tr;
e7e09b49
LLL
3456
3457 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3458 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6867a90b 3459 assert(mb_y || linesize <= block_offset[i]);
e7e09b49
LLL
3460 if(!topright_avail){
3461 tr= ptr[3 - linesize]*0x01010101;
3462 topright= (uint8_t*) &tr;
115329f1 3463 }else
e7e09b49 3464 topright= ptr + 4 - linesize;
a9799653 3465 }else
e7e09b49
LLL
3466 topright= NULL;
3467
3468 h->pred4x4[ dir ](ptr, topright, linesize);
ef9d1d15
LM
3469 nnz = h->non_zero_count_cache[ scan8[i] ];
3470 if(nnz){
3471 if(s->codec_id == CODEC_ID_H264){
3472 if(nnz == 1 && h->mb[i*16])
3473 idct_dc_add(ptr, h->mb + i*16, linesize);
3474 else
3475 idct_add(ptr, h->mb + i*16, linesize);
3476 }else
e7e09b49
LLL
3477 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3478 }
8b82a956 3479 }
0da71265 3480 }
e7e09b49
LLL
3481 }else{
3482 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
36940eca
LM
3483 if(s->codec_id == CODEC_ID_H264){
3484 if(!transform_bypass)
239ea04c 3485 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
36940eca 3486 }else
e7e09b49 3487 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
0da71265 3488 }
6ba71fc4
LLL
3489 if(h->deblocking_filter) {
3490 if (h->mb_aff_frame) {
3491 if (bottom) {
3492 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3493 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3494 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3495 s->mb_y--;
3496 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3497 s->mb_y++;
3498 }
3499 } else {
3500 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3501