cosmetics (initalize variables where they are used)
[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);
115329f1 401
0da71265
MN
402 w *= size;
403 stride *= size;
115329f1 404
4733abcb 405 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
d9c780a8 406 assert((stride&(w-1))==0);
3b66c4c5 407//FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
0da71265
MN
408 if(w==2 && h==2){
409 *(uint16_t*)(p + 0)=
410 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
411 }else if(w==2 && h==4){
412 *(uint16_t*)(p + 0*stride)=
413 *(uint16_t*)(p + 1*stride)=
414 *(uint16_t*)(p + 2*stride)=
415 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
f7a8c179
MN
416 }else if(w==4 && h==1){
417 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
0da71265
MN
418 }else if(w==4 && h==2){
419 *(uint32_t*)(p + 0*stride)=
420 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
421 }else if(w==4 && h==4){
422 *(uint32_t*)(p + 0*stride)=
423 *(uint32_t*)(p + 1*stride)=
424 *(uint32_t*)(p + 2*stride)=
425 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
426 }else if(w==8 && h==1){
427 *(uint32_t*)(p + 0)=
428 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
429 }else if(w==8 && h==2){
430 *(uint32_t*)(p + 0 + 0*stride)=
431 *(uint32_t*)(p + 4 + 0*stride)=
432 *(uint32_t*)(p + 0 + 1*stride)=
433 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
434 }else if(w==8 && h==4){
435 *(uint64_t*)(p + 0*stride)=
436 *(uint64_t*)(p + 1*stride)=
437 *(uint64_t*)(p + 2*stride)=
438 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
439 }else if(w==16 && h==2){
440 *(uint64_t*)(p + 0+0*stride)=
441 *(uint64_t*)(p + 8+0*stride)=
442 *(uint64_t*)(p + 0+1*stride)=
443 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
444 }else if(w==16 && h==4){
445 *(uint64_t*)(p + 0+0*stride)=
446 *(uint64_t*)(p + 8+0*stride)=
447 *(uint64_t*)(p + 0+1*stride)=
448 *(uint64_t*)(p + 8+1*stride)=
449 *(uint64_t*)(p + 0+2*stride)=
450 *(uint64_t*)(p + 8+2*stride)=
451 *(uint64_t*)(p + 0+3*stride)=
452 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
453 }else
454 assert(0);
455}
456
70abb407 457static void fill_caches(H264Context *h, int mb_type, int for_deblock){
0da71265 458 MpegEncContext * const s = &h->s;
7bc9090a 459 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
460 int topleft_xy, top_xy, topright_xy, left_xy[2];
461 int topleft_type, top_type, topright_type, left_type[2];
6867a90b 462 int left_block[8];
0da71265
MN
463 int i;
464
e2e5894a
LM
465 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
466 // the actual condition is whether we're on the edge of a slice,
467 // and even then the intra and nnz parts are unnecessary.
468 if(for_deblock && h->slice_num == 1)
469 return;
470
115329f1
DB
471 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
472
6867a90b
LLL
473 top_xy = mb_xy - s->mb_stride;
474 topleft_xy = top_xy - 1;
475 topright_xy= top_xy + 1;
476 left_xy[1] = left_xy[0] = mb_xy-1;
477 left_block[0]= 0;
478 left_block[1]= 1;
479 left_block[2]= 2;
480 left_block[3]= 3;
481 left_block[4]= 7;
482 left_block[5]= 10;
483 left_block[6]= 8;
484 left_block[7]= 11;
6ba71fc4 485 if(h->mb_aff_frame){
6867a90b
LLL
486 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
487 const int top_pair_xy = pair_xy - s->mb_stride;
488 const int topleft_pair_xy = top_pair_xy - 1;
489 const int topright_pair_xy = top_pair_xy + 1;
490 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
491 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
492 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
493 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
494 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
495 const int bottom = (s->mb_y & 1);
496 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);
497 if (bottom
498 ? !curr_mb_frame_flag // bottom macroblock
499 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
500 ) {
501 top_xy -= s->mb_stride;
502 }
503 if (bottom
504 ? !curr_mb_frame_flag // bottom macroblock
505 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
506 ) {
507 topleft_xy -= s->mb_stride;
508 }
509 if (bottom
510 ? !curr_mb_frame_flag // bottom macroblock
511 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
512 ) {
513 topright_xy -= s->mb_stride;
514 }
515 if (left_mb_frame_flag != curr_mb_frame_flag) {
516 left_xy[1] = left_xy[0] = pair_xy - 1;
517 if (curr_mb_frame_flag) {
518 if (bottom) {
519 left_block[0]= 2;
520 left_block[1]= 2;
521 left_block[2]= 3;
522 left_block[3]= 3;
523 left_block[4]= 8;
524 left_block[5]= 11;
525 left_block[6]= 8;
526 left_block[7]= 11;
527 } else {
528 left_block[0]= 0;
529 left_block[1]= 0;
530 left_block[2]= 1;
531 left_block[3]= 1;
532 left_block[4]= 7;
533 left_block[5]= 10;
534 left_block[6]= 7;
535 left_block[7]= 10;
536 }
537 } else {
538 left_xy[1] += s->mb_stride;
539 //left_block[0]= 0;
540 left_block[1]= 2;
541 left_block[2]= 0;
542 left_block[3]= 2;
543 //left_block[4]= 7;
544 left_block[5]= 10;
545 left_block[6]= 7;
546 left_block[7]= 10;
547 }
548 }
0da71265
MN
549 }
550
826de46e
LLL
551 h->top_mb_xy = top_xy;
552 h->left_mb_xy[0] = left_xy[0];
553 h->left_mb_xy[1] = left_xy[1];
6ba71fc4 554 if(for_deblock){
46f2f05f
MN
555 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
556 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
557 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
558 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
559 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
560 }else{
561 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
562 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
563 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
564 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
565 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
566 }
0da71265
MN
567
568 if(IS_INTRA(mb_type)){
115329f1
DB
569 h->topleft_samples_available=
570 h->top_samples_available=
0da71265
MN
571 h->left_samples_available= 0xFFFF;
572 h->topright_samples_available= 0xEEEA;
573
574 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
575 h->topleft_samples_available= 0xB3FF;
576 h->top_samples_available= 0x33FF;
577 h->topright_samples_available= 0x26EA;
578 }
579 for(i=0; i<2; i++){
580 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
581 h->topleft_samples_available&= 0xDF5F;
582 h->left_samples_available&= 0x5F5F;
583 }
584 }
115329f1 585
0da71265
MN
586 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
587 h->topleft_samples_available&= 0x7FFF;
115329f1 588
0da71265
MN
589 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
590 h->topright_samples_available&= 0xFBFF;
115329f1 591
0da71265
MN
592 if(IS_INTRA4x4(mb_type)){
593 if(IS_INTRA4x4(top_type)){
594 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
595 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
596 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
597 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
598 }else{
599 int pred;
6fbcaaa0 600 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
0da71265 601 pred= -1;
6fbcaaa0
LLL
602 else{
603 pred= 2;
0da71265
MN
604 }
605 h->intra4x4_pred_mode_cache[4+8*0]=
606 h->intra4x4_pred_mode_cache[5+8*0]=
607 h->intra4x4_pred_mode_cache[6+8*0]=
608 h->intra4x4_pred_mode_cache[7+8*0]= pred;
609 }
610 for(i=0; i<2; i++){
611 if(IS_INTRA4x4(left_type[i])){
612 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
613 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
614 }else{
615 int pred;
6fbcaaa0 616 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
0da71265 617 pred= -1;
6fbcaaa0
LLL
618 else{
619 pred= 2;
0da71265
MN
620 }
621 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
622 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
623 }
624 }
625 }
626 }
115329f1
DB
627
628
0da71265 629/*
115329f1
DB
6300 . T T. T T T T
6311 L . .L . . . .
6322 L . .L . . . .
6333 . T TL . . . .
6344 L . .L . . . .
6355 L . .. . . . .
0da71265
MN
636*/
637//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
638 if(top_type){
6867a90b
LLL
639 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
640 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
641 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
53c05b1e 642 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
115329f1 643
6867a90b 644 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
53c05b1e 645 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
115329f1 646
6867a90b 647 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
53c05b1e 648 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
115329f1 649
0da71265 650 }else{
115329f1 651 h->non_zero_count_cache[4+8*0]=
0da71265
MN
652 h->non_zero_count_cache[5+8*0]=
653 h->non_zero_count_cache[6+8*0]=
654 h->non_zero_count_cache[7+8*0]=
115329f1 655
0da71265
MN
656 h->non_zero_count_cache[1+8*0]=
657 h->non_zero_count_cache[2+8*0]=
115329f1 658
0da71265 659 h->non_zero_count_cache[1+8*3]=
3981c385 660 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
115329f1 661
0da71265 662 }
826de46e 663
6867a90b
LLL
664 for (i=0; i<2; i++) {
665 if(left_type[i]){
666 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
667 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
668 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
669 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
6867a90b 670 }else{
115329f1
DB
671 h->non_zero_count_cache[3+8*1 + 2*8*i]=
672 h->non_zero_count_cache[3+8*2 + 2*8*i]=
673 h->non_zero_count_cache[0+8*1 + 8*i]=
6867a90b 674 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
826de46e
LLL
675 }
676 }
677
678 if( h->pps.cabac ) {
679 // top_cbp
680 if(top_type) {
681 h->top_cbp = h->cbp_table[top_xy];
682 } else if(IS_INTRA(mb_type)) {
683 h->top_cbp = 0x1C0;
684 } else {
685 h->top_cbp = 0;
686 }
687 // left_cbp
688 if (left_type[0]) {
689 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
690 } else if(IS_INTRA(mb_type)) {
691 h->left_cbp = 0x1C0;
692 } else {
693 h->left_cbp = 0;
694 }
695 if (left_type[0]) {
696 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
697 }
698 if (left_type[1]) {
699 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
6867a90b 700 }
0da71265 701 }
6867a90b 702
0da71265 703#if 1
5ad984c9 704 //FIXME direct mb can skip much of this
e2e5894a 705 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
0da71265 706 int list;
e2e5894a
LM
707 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
708 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
0da71265
MN
709 /*if(!h->mv_cache_clean[list]){
710 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
711 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
712 h->mv_cache_clean[list]= 1;
713 }*/
5ad984c9 714 continue;
0da71265
MN
715 }
716 h->mv_cache_clean[list]= 0;
115329f1 717
53b19144 718 if(USES_LIST(top_type, list)){
0da71265
MN
719 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
720 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
721 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
722 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
723 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
724 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
725 h->ref_cache[list][scan8[0] + 0 - 1*8]=
726 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
727 h->ref_cache[list][scan8[0] + 2 - 1*8]=
728 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
729 }else{
115329f1
DB
730 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
731 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
732 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
0da71265
MN
733 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
734 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
735 }
736
0da71265 737 //FIXME unify cleanup or sth
53b19144 738 if(USES_LIST(left_type[0], list)){
0da71265
MN
739 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
740 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
741 *(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]];
742 *(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 743 h->ref_cache[list][scan8[0] - 1 + 0*8]=
0da71265
MN
744 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
745 }else{
746 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
747 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
748 h->ref_cache[list][scan8[0] - 1 + 0*8]=
749 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
750 }
115329f1 751
53b19144 752 if(USES_LIST(left_type[1], list)){
0da71265
MN
753 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
754 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
755 *(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]];
756 *(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 757 h->ref_cache[list][scan8[0] - 1 + 2*8]=
0da71265
MN
758 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
759 }else{
760 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
761 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
762 h->ref_cache[list][scan8[0] - 1 + 2*8]=
763 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
d9c780a8 764 assert((!left_type[0]) == (!left_type[1]));
0da71265
MN
765 }
766
e2e5894a 767 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
46f2f05f
MN
768 continue;
769
53b19144 770 if(USES_LIST(topleft_type, list)){
e2e5894a
LM
771 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
772 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
773 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
774 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
775 }else{
776 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
777 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
778 }
115329f1 779
53b19144 780 if(USES_LIST(topright_type, list)){
e2e5894a
LM
781 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
782 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
783 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
784 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
785 }else{
786 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
787 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
788 }
e2e5894a 789
115329f1
DB
790
791 h->ref_cache[list][scan8[5 ]+1] =
792 h->ref_cache[list][scan8[7 ]+1] =
3b66c4c5 793 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
115329f1 794 h->ref_cache[list][scan8[4 ]] =
0da71265
MN
795 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
796 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
797 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
3b66c4c5 798 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
0da71265
MN
799 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
800 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
9e528114
LA
801
802 if( h->pps.cabac ) {
803 /* XXX beurk, Load mvd */
53b19144 804 if(USES_LIST(topleft_type, list)){
9e528114
LA
805 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
806 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
807 }else{
808 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
809 }
810
53b19144 811 if(USES_LIST(top_type, list)){
9e528114
LA
812 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
813 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
814 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
815 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
816 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
817 }else{
115329f1
DB
818 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
819 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
820 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
9e528114
LA
821 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
822 }
53b19144 823 if(USES_LIST(left_type[0], list)){
9e528114
LA
824 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
825 *(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]];
826 *(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]];
827 }else{
828 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
829 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
830 }
53b19144 831 if(USES_LIST(left_type[1], list)){
9e528114
LA
832 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
833 *(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]];
834 *(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]];
835 }else{
836 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
837 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
838 }
839 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
840 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
3b66c4c5 841 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
9e528114
LA
842 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
843 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
5ad984c9
LM
844
845 if(h->slice_type == B_TYPE){
846 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
847
848 if(IS_DIRECT(top_type)){
849 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
850 }else if(IS_8X8(top_type)){
851 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
852 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
853 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
854 }else{
855 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
856 }
115329f1 857
5ad984c9
LM
858 //FIXME interlacing
859 if(IS_DIRECT(left_type[0])){
860 h->direct_cache[scan8[0] - 1 + 0*8]=
861 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
862 }else if(IS_8X8(left_type[0])){
863 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
864 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
865 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
866 }else{
867 h->direct_cache[scan8[0] - 1 + 0*8]=
868 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
869 }
870 }
9e528114 871 }
0da71265 872 }
0da71265
MN
873 }
874#endif
43efd19a
LM
875
876 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
0da71265
MN
877}
878
879static inline void write_back_intra_pred_mode(H264Context *h){
880 MpegEncContext * const s = &h->s;
7bc9090a 881 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
882
883 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
884 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
885 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
886 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
887 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
888 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
889 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
890}
891
892/**
893 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
894 */
895static inline int check_intra4x4_pred_mode(H264Context *h){
896 MpegEncContext * const s = &h->s;
897 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
898 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
899 int i;
115329f1 900
0da71265
MN
901 if(!(h->top_samples_available&0x8000)){
902 for(i=0; i<4; i++){
903 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
904 if(status<0){
9b879566 905 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
906 return -1;
907 } else if(status){
908 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
909 }
910 }
911 }
115329f1 912
0da71265
MN
913 if(!(h->left_samples_available&0x8000)){
914 for(i=0; i<4; i++){
915 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
916 if(status<0){
9b879566 917 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
918 return -1;
919 } else if(status){
920 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
921 }
922 }
923 }
924
925 return 0;
926} //FIXME cleanup like next
927
928/**
929 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
930 */
931static inline int check_intra_pred_mode(H264Context *h, int mode){
932 MpegEncContext * const s = &h->s;
933 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
934 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
115329f1 935
5175b937
LLL
936 if(mode < 0 || mode > 6) {
937 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 938 return -1;
5175b937 939 }
115329f1 940
0da71265
MN
941 if(!(h->top_samples_available&0x8000)){
942 mode= top[ mode ];
943 if(mode<0){
9b879566 944 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
945 return -1;
946 }
947 }
115329f1 948
0da71265
MN
949 if(!(h->left_samples_available&0x8000)){
950 mode= left[ mode ];
951 if(mode<0){
9b879566 952 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 953 return -1;
115329f1 954 }
0da71265
MN
955 }
956
957 return mode;
958}
959
960/**
961 * gets the predicted intra4x4 prediction mode.
962 */
963static inline int pred_intra_mode(H264Context *h, int n){
964 const int index8= scan8[n];
965 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
966 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
967 const int min= FFMIN(left, top);
968
95c26348 969 tprintf("mode:%d %d min:%d\n", left ,top, min);
0da71265
MN
970
971 if(min<0) return DC_PRED;
972 else return min;
973}
974
975static inline void write_back_non_zero_count(H264Context *h){
976 MpegEncContext * const s = &h->s;
7bc9090a 977 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 978
6867a90b
LLL
979 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
980 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
981 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
53c05b1e 982 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
6867a90b
LLL
983 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
984 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
985 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
115329f1 986
6867a90b 987 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
53c05b1e 988 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
6867a90b 989 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
53c05b1e 990
6867a90b 991 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
53c05b1e 992 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
6867a90b 993 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
0da71265
MN
994}
995
996/**
997 * gets the predicted number of non zero coefficients.
998 * @param n block index
999 */
1000static inline int pred_non_zero_count(H264Context *h, int n){
1001 const int index8= scan8[n];
1002 const int left= h->non_zero_count_cache[index8 - 1];
1003 const int top = h->non_zero_count_cache[index8 - 8];
1004 int i= left + top;
115329f1 1005
0da71265
MN
1006 if(i<64) i= (i+1)>>1;
1007
95c26348 1008 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
0da71265
MN
1009
1010 return i&31;
1011}
1012
1924f3ce
MN
1013static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1014 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1015
1016 if(topright_ref != PART_NOT_AVAILABLE){
1017 *C= h->mv_cache[list][ i - 8 + part_width ];
1018 return topright_ref;
1019 }else{
95c26348
MN
1020 tprintf("topright MV not available\n");
1021
1924f3ce
MN
1022 *C= h->mv_cache[list][ i - 8 - 1 ];
1023 return h->ref_cache[list][ i - 8 - 1 ];
1024 }
1025}
1026
0da71265
MN
1027/**
1028 * gets the predicted MV.
1029 * @param n the block index
1030 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1031 * @param mx the x component of the predicted motion vector
1032 * @param my the y component of the predicted motion vector
1033 */
1034static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
0da71265
MN
1035 const int index8= scan8[n];
1036 const int top_ref= h->ref_cache[list][ index8 - 8 ];
0da71265
MN
1037 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1038 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1039 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1924f3ce
MN
1040 const int16_t * C;
1041 int diagonal_ref, match_count;
1042
0da71265 1043 assert(part_width==1 || part_width==2 || part_width==4);
1924f3ce 1044
0da71265 1045/* mv_cache
115329f1 1046 B . . A T T T T
0da71265
MN
1047 U . . L . . , .
1048 U . . L . . . .
1049 U . . L . . , .
1050 . . . L . . . .
1051*/
1924f3ce
MN
1052
1053 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1054 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
827c91bf 1055 tprintf("pred_motion match_count=%d\n", match_count);
1924f3ce
MN
1056 if(match_count > 1){ //most common
1057 *mx= mid_pred(A[0], B[0], C[0]);
1058 *my= mid_pred(A[1], B[1], C[1]);
1059 }else if(match_count==1){
1060 if(left_ref==ref){
1061 *mx= A[0];
115329f1 1062 *my= A[1];
1924f3ce
MN
1063 }else if(top_ref==ref){
1064 *mx= B[0];
115329f1 1065 *my= B[1];
0da71265 1066 }else{
1924f3ce 1067 *mx= C[0];
115329f1 1068 *my= C[1];
0da71265
MN
1069 }
1070 }else{
1924f3ce 1071 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
0da71265 1072 *mx= A[0];
115329f1 1073 *my= A[1];
0da71265 1074 }else{
1924f3ce
MN
1075 *mx= mid_pred(A[0], B[0], C[0]);
1076 *my= mid_pred(A[1], B[1], C[1]);
0da71265 1077 }
0da71265 1078 }
115329f1 1079
af6e2fed 1080 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
1081}
1082
1083/**
1084 * gets the directionally predicted 16x8 MV.
1085 * @param n the block index
1086 * @param mx the x component of the predicted motion vector
1087 * @param my the y component of the predicted motion vector
1088 */
1089static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1090 if(n==0){
1091 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1092 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1093
579448a0 1094 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 1095
0da71265
MN
1096 if(top_ref == ref){
1097 *mx= B[0];
1098 *my= B[1];
1099 return;
1100 }
1101 }else{
1102 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1103 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
115329f1 1104
579448a0 1105 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
1106
1107 if(left_ref == ref){
1108 *mx= A[0];
1109 *my= A[1];
1110 return;
1111 }
1112 }
1113
1114 //RARE
1115 pred_motion(h, n, 4, list, ref, mx, my);
1116}
1117
1118/**
1119 * gets the directionally predicted 8x16 MV.
1120 * @param n the block index
1121 * @param mx the x component of the predicted motion vector
1122 * @param my the y component of the predicted motion vector
1123 */
1124static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1125 if(n==0){
1126 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1127 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
115329f1 1128
579448a0 1129 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
1130
1131 if(left_ref == ref){
1132 *mx= A[0];
1133 *my= A[1];
1134 return;
1135 }
1136 }else{
1924f3ce
MN
1137 const int16_t * C;
1138 int diagonal_ref;
1139
1140 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
115329f1 1141
579448a0 1142 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 1143
115329f1 1144 if(diagonal_ref == ref){
0da71265
MN
1145 *mx= C[0];
1146 *my= C[1];
1147 return;
1148 }
0da71265
MN
1149 }
1150
1151 //RARE
1152 pred_motion(h, n, 2, list, ref, mx, my);
1153}
1154
1155static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
0da71265
MN
1156 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1157 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1158
5175b937 1159 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
0da71265
MN
1160
1161 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1162 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1163 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
115329f1 1164
0da71265
MN
1165 *mx = *my = 0;
1166 return;
1167 }
115329f1 1168
0da71265
MN
1169 pred_motion(h, 0, 4, 0, 0, mx, my);
1170
1171 return;
1172}
1173
5ad984c9
LM
1174static inline void direct_dist_scale_factor(H264Context * const h){
1175 const int poc = h->s.current_picture_ptr->poc;
1176 const int poc1 = h->ref_list[1][0].poc;
1177 int i;
1178 for(i=0; i<h->ref_count[0]; i++){
1179 int poc0 = h->ref_list[0][i].poc;
1180 int td = clip(poc1 - poc0, -128, 127);
1181 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1182 h->dist_scale_factor[i] = 256;
1183 }else{
1184 int tb = clip(poc - poc0, -128, 127);
1185 int tx = (16384 + (ABS(td) >> 1)) / td;
1186 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1187 }
1188 }
1189}
2f944356
LM
1190static inline void direct_ref_list_init(H264Context * const h){
1191 MpegEncContext * const s = &h->s;
1192 Picture * const ref1 = &h->ref_list[1][0];
1193 Picture * const cur = s->current_picture_ptr;
1194 int list, i, j;
1195 if(cur->pict_type == I_TYPE)
1196 cur->ref_count[0] = 0;
1197 if(cur->pict_type != B_TYPE)
1198 cur->ref_count[1] = 0;
1199 for(list=0; list<2; list++){
1200 cur->ref_count[list] = h->ref_count[list];
1201 for(j=0; j<h->ref_count[list]; j++)
1202 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1203 }
1204 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1205 return;
1206 for(list=0; list<2; list++){
1207 for(i=0; i<ref1->ref_count[list]; i++){
1208 const int poc = ref1->ref_poc[list][i];
171c4076 1209 h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
2f944356
LM
1210 for(j=0; j<h->ref_count[list]; j++)
1211 if(h->ref_list[list][j].poc == poc){
1212 h->map_col_to_list0[list][i] = j;
1213 break;
1214 }
1215 }
1216 }
1217}
5ad984c9
LM
1218
1219static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1220 MpegEncContext * const s = &h->s;
1221 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1222 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1223 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1224 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1225 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
4866bd2b 1226 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
5ad984c9 1227 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
2f944356 1228 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
5ad984c9
LM
1229 const int is_b8x8 = IS_8X8(*mb_type);
1230 int sub_mb_type;
1231 int i8, i4;
1232
1233 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1234 /* FIXME save sub mb types from previous frames (or derive from MVs)
1235 * so we know exactly what block size to use */
1236 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
3622988f 1237 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1238 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1239 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1240 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1241 }else{
1242 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
3622988f 1243 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1244 }
1245 if(!is_b8x8)
1246 *mb_type |= MB_TYPE_DIRECT2;
1247
827c91bf 1248 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 1249
5ad984c9
LM
1250 if(h->direct_spatial_mv_pred){
1251 int ref[2];
1252 int mv[2][2];
1253 int list;
1254
1255 /* ref = min(neighbors) */
1256 for(list=0; list<2; list++){
1257 int refa = h->ref_cache[list][scan8[0] - 1];
1258 int refb = h->ref_cache[list][scan8[0] - 8];
1259 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1260 if(refc == -2)
1261 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1262 ref[list] = refa;
1263 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1264 ref[list] = refb;
1265 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1266 ref[list] = refc;
1267 if(ref[list] < 0)
1268 ref[list] = -1;
1269 }
1270
1271 if(ref[0] < 0 && ref[1] < 0){
1272 ref[0] = ref[1] = 0;
1273 mv[0][0] = mv[0][1] =
1274 mv[1][0] = mv[1][1] = 0;
1275 }else{
1276 for(list=0; list<2; list++){
1277 if(ref[list] >= 0)
1278 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1279 else
1280 mv[list][0] = mv[list][1] = 0;
1281 }
1282 }
1283
1284 if(ref[1] < 0){
1285 *mb_type &= ~MB_TYPE_P0L1;
1286 sub_mb_type &= ~MB_TYPE_P0L1;
1287 }else if(ref[0] < 0){
1288 *mb_type &= ~MB_TYPE_P0L0;
1289 sub_mb_type &= ~MB_TYPE_P0L0;
1290 }
1291
1292 if(IS_16X16(*mb_type)){
cec93959
LM
1293 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1294 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
115329f1 1295 if(!IS_INTRA(mb_type_col)
bf4e3bd2
MR
1296 && ( (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1297 || (l1ref0[0] < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1298 && (h->x264_build>33 || !h->x264_build)))){
5ad984c9
LM
1299 if(ref[0] > 0)
1300 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1301 else
1302 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1303 if(ref[1] > 0)
1304 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1305 else
1306 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1307 }else{
1308 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1309 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1310 }
1311 }else{
1312 for(i8=0; i8<4; i8++){
1313 const int x8 = i8&1;
1314 const int y8 = i8>>1;
115329f1 1315
5ad984c9
LM
1316 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1317 continue;
1318 h->sub_mb_type[i8] = sub_mb_type;
115329f1 1319
5ad984c9
LM
1320 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1321 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
cec93959
LM
1322 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1323 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
115329f1 1324
5ad984c9 1325 /* col_zero_flag */
115329f1
DB
1326 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1327 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
bf4e3bd2 1328 && (h->x264_build>33 || !h->x264_build)))){
4866bd2b 1329 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
f1f17e54
LM
1330 if(IS_SUB_8X8(sub_mb_type)){
1331 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1332 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1333 if(ref[0] == 0)
1334 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1335 if(ref[1] == 0)
1336 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1337 }
1338 }else
5ad984c9 1339 for(i4=0; i4<4; i4++){
4866bd2b 1340 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9
LM
1341 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1342 if(ref[0] == 0)
1343 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1344 if(ref[1] == 0)
1345 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1346 }
1347 }
1348 }
1349 }
1350 }
1351 }else{ /* direct temporal mv pred */
5ad984c9
LM
1352 if(IS_16X16(*mb_type)){
1353 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1354 if(IS_INTRA(mb_type_col)){
1355 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1356 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1357 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1358 }else{
2f944356
LM
1359 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1360 : h->map_col_to_list0[1][l1ref1[0]];
5ad984c9 1361 const int dist_scale_factor = h->dist_scale_factor[ref0];
8583bef8 1362 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
5ad984c9
LM
1363 int mv_l0[2];
1364 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1365 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1366 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1367 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1368 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);
1369 }
1370 }else{
1371 for(i8=0; i8<4; i8++){
1372 const int x8 = i8&1;
1373 const int y8 = i8>>1;
1374 int ref0, dist_scale_factor;
bf4e3bd2 1375 const int16_t (*l1mv)[2]= l1mv0;
8583bef8 1376
5ad984c9
LM
1377 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1378 continue;
1379 h->sub_mb_type[i8] = sub_mb_type;
1380 if(IS_INTRA(mb_type_col)){
1381 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1382 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1383 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1384 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1385 continue;
1386 }
115329f1 1387
5ad984c9 1388 ref0 = l1ref0[x8 + y8*h->b8_stride];
2f944356
LM
1389 if(ref0 >= 0)
1390 ref0 = h->map_col_to_list0[0][ref0];
8583bef8 1391 else{
2f944356 1392 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
8583bef8
MN
1393 l1mv= l1mv1;
1394 }
5ad984c9 1395 dist_scale_factor = h->dist_scale_factor[ref0];
115329f1 1396
5ad984c9
LM
1397 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1398 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
f1f17e54
LM
1399 if(IS_SUB_8X8(sub_mb_type)){
1400 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1401 int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
1402 int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
1403 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1404 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1405 }else
5ad984c9 1406 for(i4=0; i4<4; i4++){
8583bef8 1407 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
5ad984c9
LM
1408 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1409 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1410 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1411 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1412 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1413 }
1414 }
1415 }
1416 }
1417}
1418
0da71265
MN
1419static inline void write_back_motion(H264Context *h, int mb_type){
1420 MpegEncContext * const s = &h->s;
0da71265
MN
1421 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1422 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1423 int list;
1424
2ea39252
LM
1425 if(!USES_LIST(mb_type, 0))
1426 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1427
0da71265
MN
1428 for(list=0; list<2; list++){
1429 int y;
53b19144 1430 if(!USES_LIST(mb_type, list))
5ad984c9 1431 continue;
115329f1 1432
0da71265
MN
1433 for(y=0; y<4; y++){
1434 *(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];
1435 *(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];
1436 }
9e528114
LA
1437 if( h->pps.cabac ) {
1438 for(y=0; y<4; y++){
1439 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1440 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1441 }
1442 }
53b19144
LM
1443
1444 {
1445 uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1446 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1447 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1448 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1449 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
0da71265
MN
1450 }
1451 }
115329f1 1452
5ad984c9
LM
1453 if(h->slice_type == B_TYPE && h->pps.cabac){
1454 if(IS_8X8(mb_type)){
53b19144
LM
1455 uint8_t *direct_table = &h->direct_table[b8_xy];
1456 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1457 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1458 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
5ad984c9
LM
1459 }
1460 }
0da71265
MN
1461}
1462
1463/**
1464 * Decodes a network abstraction layer unit.
1465 * @param consumed is the number of bytes used as input
1466 * @param length is the length of the array
3b66c4c5 1467 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
115329f1 1468 * @returns decoded bytes, might be src+1 if no escapes
0da71265
MN
1469 */
1470static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1471 int i, si, di;
1472 uint8_t *dst;
1473
bb270c08 1474// src[0]&0x80; //forbidden bit
0da71265
MN
1475 h->nal_ref_idc= src[0]>>5;
1476 h->nal_unit_type= src[0]&0x1F;
1477
1478 src++; length--;
115329f1 1479#if 0
0da71265
MN
1480 for(i=0; i<length; i++)
1481 printf("%2X ", src[i]);
1482#endif
1483 for(i=0; i+1<length; i+=2){
1484 if(src[i]) continue;
1485 if(i>0 && src[i-1]==0) i--;
1486 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1487 if(src[i+2]!=3){
1488 /* startcode, so we must be past the end */
1489 length=i;
1490 }
1491 break;
1492 }
1493 }
1494
1495 if(i>=length-1){ //no escaped 0
1496 *dst_length= length;
1497 *consumed= length+1; //+1 for the header
115329f1 1498 return src;
0da71265
MN
1499 }
1500
1501 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1502 dst= h->rbsp_buffer;
1503
3b66c4c5 1504//printf("decoding esc\n");
0da71265 1505 si=di=0;
115329f1 1506 while(si<length){
0da71265
MN
1507 //remove escapes (very rare 1:2^22)
1508 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1509 if(src[si+2]==3){ //escape
1510 dst[di++]= 0;
1511 dst[di++]= 0;
1512 si+=3;
c8470cc1 1513 continue;
0da71265
MN
1514 }else //next start code
1515 break;
1516 }
1517
1518 dst[di++]= src[si++];
1519 }
1520
1521 *dst_length= di;
1522 *consumed= si + 1;//+1 for the header
1523//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1524 return dst;
1525}
1526
e5017ab8 1527#if 0
0da71265
MN
1528/**
1529 * @param src the data which should be escaped
1530 * @param dst the target buffer, dst+1 == src is allowed as a special case
1531 * @param length the length of the src data
1532 * @param dst_length the length of the dst array
1533 * @returns length of escaped data in bytes or -1 if an error occured
1534 */
1535static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1536 int i, escape_count, si, di;
1537 uint8_t *temp;
115329f1 1538
0da71265
MN
1539 assert(length>=0);
1540 assert(dst_length>0);
115329f1 1541
0da71265
MN
1542 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1543
1544 if(length==0) return 1;
1545
1546 escape_count= 0;
1547 for(i=0; i<length; i+=2){
1548 if(src[i]) continue;
115329f1 1549 if(i>0 && src[i-1]==0)
0da71265
MN
1550 i--;
1551 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1552 escape_count++;
1553 i+=2;
1554 }
1555 }
115329f1
DB
1556
1557 if(escape_count==0){
0da71265
MN
1558 if(dst+1 != src)
1559 memcpy(dst+1, src, length);
1560 return length + 1;
1561 }
115329f1 1562
0da71265
MN
1563 if(length + escape_count + 1> dst_length)
1564 return -1;
1565
1566 //this should be damn rare (hopefully)
1567
1568 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1569 temp= h->rbsp_buffer;
1570//printf("encoding esc\n");
115329f1 1571
0da71265
MN
1572 si= 0;
1573 di= 0;
1574 while(si < length){
1575 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1576 temp[di++]= 0; si++;
1577 temp[di++]= 0; si++;
115329f1 1578 temp[di++]= 3;
0da71265
MN
1579 temp[di++]= src[si++];
1580 }
1581 else
1582 temp[di++]= src[si++];
1583 }
1584 memcpy(dst+1, temp, length+escape_count);
115329f1 1585
0da71265 1586 assert(di == length+escape_count);
115329f1 1587
0da71265
MN
1588 return di + 1;
1589}
1590
1591/**
1592 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1593 */
1594static void encode_rbsp_trailing(PutBitContext *pb){
1595 int length;
1596 put_bits(pb, 1, 1);
fe455f33 1597 length= (-put_bits_count(pb))&7;
0da71265
MN
1598 if(length) put_bits(pb, length, 0);
1599}
e5017ab8 1600#endif
0da71265
MN
1601
1602/**
1603 * identifies the exact end of the bitstream
1604 * @return the length of the trailing, or 0 if damaged
1605 */
1606static int decode_rbsp_trailing(uint8_t *src){
1607 int v= *src;
1608 int r;
1609
95c26348 1610 tprintf("rbsp trailing %X\n", v);
0da71265
MN
1611
1612 for(r=1; r<9; r++){
1613 if(v&1) return r;
1614 v>>=1;
1615 }
1616 return 0;
1617}
1618
1619/**
1620 * idct tranforms the 16 dc values and dequantize them.
1621 * @param qp quantization parameter
1622 */
239ea04c 1623static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1624#define stride 16
1625 int i;
1626 int temp[16]; //FIXME check if this is a good idea
1627 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1628 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1629
1630//memset(block, 64, 2*256);
1631//return;
1632 for(i=0; i<4; i++){
1633 const int offset= y_offset[i];
1634 const int z0= block[offset+stride*0] + block[offset+stride*4];
1635 const int z1= block[offset+stride*0] - block[offset+stride*4];
1636 const int z2= block[offset+stride*1] - block[offset+stride*5];
1637 const int z3= block[offset+stride*1] + block[offset+stride*5];
1638
1639 temp[4*i+0]= z0+z3;
1640 temp[4*i+1]= z1+z2;
1641 temp[4*i+2]= z1-z2;
1642 temp[4*i+3]= z0-z3;
1643 }
1644
1645 for(i=0; i<4; i++){
1646 const int offset= x_offset[i];
1647 const int z0= temp[4*0+i] + temp[4*2+i];
1648 const int z1= temp[4*0+i] - temp[4*2+i];
1649 const int z2= temp[4*1+i] - temp[4*3+i];
1650 const int z3= temp[4*1+i] + temp[4*3+i];
1651
239ea04c
LM
1652 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1653 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1654 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1655 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
0da71265
MN
1656 }
1657}
1658
e5017ab8 1659#if 0
0da71265
MN
1660/**
1661 * dct tranforms the 16 dc values.
1662 * @param qp quantization parameter ??? FIXME
1663 */
1664static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1665// const int qmul= dequant_coeff[qp][0];
1666 int i;
1667 int temp[16]; //FIXME check if this is a good idea
1668 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1669 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1670
1671 for(i=0; i<4; i++){
1672 const int offset= y_offset[i];
1673 const int z0= block[offset+stride*0] + block[offset+stride*4];
1674 const int z1= block[offset+stride*0] - block[offset+stride*4];
1675 const int z2= block[offset+stride*1] - block[offset+stride*5];
1676 const int z3= block[offset+stride*1] + block[offset+stride*5];
1677
1678 temp[4*i+0]= z0+z3;
1679 temp[4*i+1]= z1+z2;
1680 temp[4*i+2]= z1-z2;
1681 temp[4*i+3]= z0-z3;
1682 }
1683
1684 for(i=0; i<4; i++){
1685 const int offset= x_offset[i];
1686 const int z0= temp[4*0+i] + temp[4*2+i];
1687 const int z1= temp[4*0+i] - temp[4*2+i];
1688 const int z2= temp[4*1+i] - temp[4*3+i];
1689 const int z3= temp[4*1+i] + temp[4*3+i];
1690
1691 block[stride*0 +offset]= (z0 + z3)>>1;
1692 block[stride*2 +offset]= (z1 + z2)>>1;
1693 block[stride*8 +offset]= (z1 - z2)>>1;
1694 block[stride*10+offset]= (z0 - z3)>>1;
1695 }
1696}
e5017ab8
LA
1697#endif
1698
0da71265
MN
1699#undef xStride
1700#undef stride
1701
239ea04c 1702static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1703 const int stride= 16*2;
1704 const int xStride= 16;
1705 int a,b,c,d,e;
1706
1707 a= block[stride*0 + xStride*0];
1708 b= block[stride*0 + xStride*1];
1709 c= block[stride*1 + xStride*0];
1710 d= block[stride*1 + xStride*1];
1711
1712 e= a-b;
1713 a= a+b;
1714 b= c-d;
1715 c= c+d;
1716
239ea04c
LM
1717 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1718 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1719 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1720 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
0da71265
MN
1721}
1722
e5017ab8 1723#if 0
0da71265
MN
1724static void chroma_dc_dct_c(DCTELEM *block){
1725 const int stride= 16*2;
1726 const int xStride= 16;
1727 int a,b,c,d,e;
1728
1729 a= block[stride*0 + xStride*0];
1730 b= block[stride*0 + xStride*1];
1731 c= block[stride*1 + xStride*0];
1732 d= block[stride*1 + xStride*1];
1733
1734 e= a-b;
1735 a= a+b;
1736 b= c-d;
1737 c= c+d;
1738
1739 block[stride*0 + xStride*0]= (a+c);
1740 block[stride*0 + xStride*1]= (e+b);
1741 block[stride*1 + xStride*0]= (a-c);
1742 block[stride*1 + xStride*1]= (e-b);
1743}
e5017ab8 1744#endif
0da71265
MN
1745
1746/**
1747 * gets the chroma qp.
1748 */
6ba71fc4 1749static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
115329f1 1750
6ba71fc4 1751 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
0da71265
MN
1752}
1753
1754
e5017ab8 1755#if 0
0da71265
MN
1756static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1757 int i;
1758 //FIXME try int temp instead of block
115329f1 1759
0da71265
MN
1760 for(i=0; i<4; i++){
1761 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1762 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1763 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1764 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1765 const int z0= d0 + d3;
1766 const int z3= d0 - d3;
1767 const int z1= d1 + d2;
1768 const int z2= d1 - d2;
115329f1 1769
0da71265
MN
1770 block[0 + 4*i]= z0 + z1;
1771 block[1 + 4*i]= 2*z3 + z2;
1772 block[2 + 4*i]= z0 - z1;
1773 block[3 + 4*i]= z3 - 2*z2;
115329f1 1774 }
0da71265
MN
1775
1776 for(i=0; i<4; i++){
1777 const int z0= block[0*4 + i] + block[3*4 + i];
1778 const int z3= block[0*4 + i] - block[3*4 + i];
1779 const int z1= block[1*4 + i] + block[2*4 + i];
1780 const int z2= block[1*4 + i] - block[2*4 + i];
115329f1 1781
0da71265
MN
1782 block[0*4 + i]= z0 + z1;
1783 block[1*4 + i]= 2*z3 + z2;
1784 block[2*4 + i]= z0 - z1;
1785 block[3*4 + i]= z3 - 2*z2;
1786 }
1787}
e5017ab8 1788#endif
0da71265 1789
3b66c4c5 1790//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
0da71265
MN
1791//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1792static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1793 int i;
1794 const int * const quant_table= quant_coeff[qscale];
1795 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1796 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1797 const unsigned int threshold2= (threshold1<<1);
1798 int last_non_zero;
1799
1800 if(seperate_dc){
1801 if(qscale<=18){
1802 //avoid overflows
1803 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1804 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1805 const unsigned int dc_threshold2= (dc_threshold1<<1);
1806
1807 int level= block[0]*quant_coeff[qscale+18][0];
1808 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1809 if(level>0){
1810 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1811 block[0]= level;
1812 }else{
1813 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1814 block[0]= -level;
1815 }
1816// last_non_zero = i;
1817 }else{
1818 block[0]=0;
1819 }
1820 }else{
1821 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1822 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1823 const unsigned int dc_threshold2= (dc_threshold1<<1);
1824
1825 int level= block[0]*quant_table[0];
1826 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1827 if(level>0){
1828 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1829 block[0]= level;
1830 }else{
1831 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1832 block[0]= -level;
1833 }
1834// last_non_zero = i;
1835 }else{
1836 block[0]=0;
1837 }
1838 }
1839 last_non_zero= 0;
1840 i=1;
1841 }else{
1842 last_non_zero= -1;
1843 i=0;
1844 }
1845
1846 for(; i<16; i++){
1847 const int j= scantable[i];
1848 int level= block[j]*quant_table[j];
1849
1850// if( bias+level >= (1<<(QMAT_SHIFT - 3))
1851// || bias-level >= (1<<(QMAT_SHIFT - 3))){
1852 if(((unsigned)(level+threshold1))>threshold2){
1853 if(level>0){
1854 level= (bias + level)>>QUANT_SHIFT;
1855 block[j]= level;
1856 }else{
1857 level= (bias - level)>>QUANT_SHIFT;
1858 block[j]= -level;
1859 }
1860 last_non_zero = i;
1861 }else{
1862 block[j]=0;
1863 }
1864 }
1865
1866 return last_non_zero;
1867}
1868
1869static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1870 const uint32_t a= ((uint32_t*)(src-stride))[0];
1871 ((uint32_t*)(src+0*stride))[0]= a;
1872 ((uint32_t*)(src+1*stride))[0]= a;
1873 ((uint32_t*)(src+2*stride))[0]= a;
1874 ((uint32_t*)(src+3*stride))[0]= a;
1875}
1876
1877static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1878 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1879 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1880 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1881 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1882}
1883
1884static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1885 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1886 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
115329f1
DB
1887
1888 ((uint32_t*)(src+0*stride))[0]=
1889 ((uint32_t*)(src+1*stride))[0]=
1890 ((uint32_t*)(src+2*stride))[0]=
1891 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1892}
1893
1894static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1895 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
115329f1
DB
1896
1897 ((uint32_t*)(src+0*stride))[0]=
1898 ((uint32_t*)(src+1*stride))[0]=
1899 ((uint32_t*)(src+2*stride))[0]=
1900 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1901}
1902
1903static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1904 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
115329f1
DB
1905
1906 ((uint32_t*)(src+0*stride))[0]=
1907 ((uint32_t*)(src+1*stride))[0]=
1908 ((uint32_t*)(src+2*stride))[0]=
1909 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1910}
1911
1912static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1913 ((uint32_t*)(src+0*stride))[0]=
1914 ((uint32_t*)(src+1*stride))[0]=
1915 ((uint32_t*)(src+2*stride))[0]=
0da71265
MN
1916 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1917}
1918
1919
1920#define LOAD_TOP_RIGHT_EDGE\
1921 const int t4= topright[0];\
1922 const int t5= topright[1];\
1923 const int t6= topright[2];\
1924 const int t7= topright[3];\
1925
1926#define LOAD_LEFT_EDGE\
1927 const int l0= src[-1+0*stride];\
1928 const int l1= src[-1+1*stride];\
1929 const int l2= src[-1+2*stride];\
1930 const int l3= src[-1+3*stride];\
1931
1932#define LOAD_TOP_EDGE\
1933 const int t0= src[ 0-1*stride];\
1934 const int t1= src[ 1-1*stride];\
1935 const int t2= src[ 2-1*stride];\
1936 const int t3= src[ 3-1*stride];\
1937
1938static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1939 const int lt= src[-1-1*stride];
1940 LOAD_TOP_EDGE
1941 LOAD_LEFT_EDGE
1942
115329f1 1943 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
0da71265 1944 src[0+2*stride]=
115329f1 1945 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
0da71265
MN
1946 src[0+1*stride]=
1947 src[1+2*stride]=
115329f1 1948 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
0da71265
MN
1949 src[0+0*stride]=
1950 src[1+1*stride]=
1951 src[2+2*stride]=
115329f1 1952 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
0da71265
MN
1953 src[1+0*stride]=
1954 src[2+1*stride]=
1955 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1956 src[2+0*stride]=
1957 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1958 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
4cfbf61b 1959}
0da71265
MN
1960
1961static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1962 LOAD_TOP_EDGE
1963 LOAD_TOP_RIGHT_EDGE
1964// LOAD_LEFT_EDGE
0da71265
MN
1965
1966 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1967 src[1+0*stride]=
1968 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1969 src[2+0*stride]=
1970 src[1+1*stride]=
1971 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1972 src[3+0*stride]=
1973 src[2+1*stride]=
1974 src[1+2*stride]=
1975 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1976 src[3+1*stride]=
1977 src[2+2*stride]=
1978 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1979 src[3+2*stride]=
1980 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1981 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
4cfbf61b 1982}
0da71265
MN
1983
1984static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1985 const int lt= src[-1-1*stride];
115329f1
DB
1986 LOAD_TOP_EDGE
1987 LOAD_LEFT_EDGE
0da71265
MN
1988 const __attribute__((unused)) int unu= l3;
1989
1990 src[0+0*stride]=
1991 src[1+2*stride]=(lt + t0 + 1)>>1;
1992 src[1+0*stride]=
1993 src[2+2*stride]=(t0 + t1 + 1)>>1;
1994 src[2+0*stride]=
1995 src[3+2*stride]=(t1 + t2 + 1)>>1;
1996 src[3+0*stride]=(t2 + t3 + 1)>>1;
1997 src[0+1*stride]=
1998 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1999 src[1+1*stride]=
2000 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2001 src[2+1*stride]=
2002 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2003 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2004 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2005 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
4cfbf61b 2006}
0da71265
MN
2007
2008static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2009 LOAD_TOP_EDGE
2010 LOAD_TOP_RIGHT_EDGE
0da71265
MN
2011 const __attribute__((unused)) int unu= t7;
2012
2013 src[0+0*stride]=(t0 + t1 + 1)>>1;
2014 src[1+0*stride]=
2015 src[0+2*stride]=(t1 + t2 + 1)>>1;
2016 src[2+0*stride]=
2017 src[1+2*stride]=(t2 + t3 + 1)>>1;
2018 src[3+0*stride]=
2019 src[2+2*stride]=(t3 + t4+ 1)>>1;
2020 src[3+2*stride]=(t4 + t5+ 1)>>1;
2021 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2022 src[1+1*stride]=
2023 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2024 src[2+1*stride]=
2025 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2026 src[3+1*stride]=
2027 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2028 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
4cfbf61b 2029}
0da71265
MN
2030
2031static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
115329f1 2032 LOAD_LEFT_EDGE
0da71265
MN
2033
2034 src[0+0*stride]=(l0 + l1 + 1)>>1;
2035 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2036 src[2+0*stride]=
2037 src[0+1*stride]=(l1 + l2 + 1)>>1;
2038 src[3+0*stride]=
2039 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2040 src[2+1*stride]=
2041 src[0+2*stride]=(l2 + l3 + 1)>>1;
2042 src[3+1*stride]=
2043 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2044 src[3+2*stride]=
2045 src[1+3*stride]=
2046 src[0+3*stride]=
2047 src[2+2*stride]=
2048 src[2+3*stride]=
2049 src[3+3*stride]=l3;
4cfbf61b 2050}
115329f1 2051
0da71265
MN
2052static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2053 const int lt= src[-1-1*stride];
115329f1
DB
2054 LOAD_TOP_EDGE
2055 LOAD_LEFT_EDGE
0da71265
MN
2056 const __attribute__((unused)) int unu= t3;
2057
2058 src[0+0*stride]=
2059 src[2+1*stride]=(lt + l0 + 1)>>1;
2060 src[1+0*stride]=
2061 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2062 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2063 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2064 src[0+1*stride]=
2065 src[2+2*stride]=(l0 + l1 + 1)>>1;
2066 src[1+1*stride]=
2067 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2068 src[0+2*stride]=
2069 src[2+3*stride]=(l1 + l2+ 1)>>1;
2070 src[1+2*stride]=
2071 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2072 src[0+3*stride]=(l2 + l3 + 1)>>1;
2073 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
4cfbf61b 2074}
0da71265
MN
2075
2076static void pred16x16_vertical_c(uint8_t *src, int stride){
2077 int i;
2078 const uint32_t a= ((uint32_t*)(src-stride))[0];
2079 const uint32_t b= ((uint32_t*)(src-stride))[1];
2080 const uint32_t c= ((uint32_t*)(src-stride))[2];
2081 const uint32_t d= ((uint32_t*)(src-stride))[3];
115329f1 2082
0da71265
MN
2083 for(i=0; i<16; i++){
2084 ((uint32_t*)(src+i*stride))[0]= a;
2085 ((uint32_t*)(src+i*stride))[1]= b;
2086 ((uint32_t*)(src+i*stride))[2]= c;
2087 ((uint32_t*)(src+i*stride))[3]= d;
2088 }
2089}
2090
2091static void pred16x16_horizontal_c(uint8_t *src, int stride){
2092 int i;
2093
2094 for(i=0; i<16; i++){
2095 ((uint32_t*)(src+i*stride))[0]=
2096 ((uint32_t*)(src+i*stride))[1]=
2097 ((uint32_t*)(src+i*stride))[2]=
2098 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2099 }
2100}
2101
2102static void pred16x16_dc_c(uint8_t *src, int stride){
2103 int i, dc=0;
2104
2105 for(i=0;i<16; i++){
2106 dc+= src[-1+i*stride];
2107 }
115329f1 2108
0da71265
MN
2109 for(i=0;i<16; i++){
2110 dc+= src[i-stride];
2111 }
2112
2113 dc= 0x01010101*((dc + 16)>>5);
2114
2115 for(i=0; i<16; i++){
2116 ((uint32_t*)(src+i*stride))[0]=
2117 ((uint32_t*)(src+i*stride))[1]=
2118 ((uint32_t*)(src+i*stride))[2]=
2119 ((uint32_t*)(src+i*stride))[3]= dc;
2120 }
2121}
2122
2123static void pred16x16_left_dc_c(uint8_t *src, int stride){
2124 int i, dc=0;
2125
2126 for(i=0;i<16; i++){
2127 dc+= src[-1+i*stride];
2128 }
115329f1 2129
0da71265
MN
2130 dc= 0x01010101*((dc + 8)>>4);
2131
2132 for(i=0; i<16; i++){
2133 ((uint32_t*)(src+i*stride))[0]=
2134 ((uint32_t*)(src+i*stride))[1]=
2135 ((uint32_t*)(src+i*stride))[2]=
2136 ((uint32_t*)(src+i*stride))[3]= dc;
2137 }
2138}
2139
2140static void pred16x16_top_dc_c(uint8_t *src, int stride){
2141 int i, dc=0;
2142
2143 for(i=0;i<16; i++){
2144 dc+= src[i-stride];
2145 }
2146 dc= 0x01010101*((dc + 8)>>4);
2147
2148 for(i=0; i<16; i++){
2149 ((uint32_t*)(src+i*stride))[0]=
2150 ((uint32_t*)(src+i*stride))[1]=
2151 ((uint32_t*)(src+i*stride))[2]=
2152 ((uint32_t*)(src+i*stride))[3]= dc;
2153 }
2154}
2155
2156static void pred16x16_128_dc_c(uint8_t *src, int stride){
2157 int i;
2158
2159 for(i=0; i<16; i++){
2160 ((uint32_t*)(src+i*stride))[0]=
2161 ((uint32_t*)(src+i*stride))[1]=
2162 ((uint32_t*)(src+i*stride))[2]=
2163 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2164 }
2165}
2166
8b82a956 2167static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
30f73fc7
MN
2168 int i, j, k;
2169 int a;
2170 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2171 const uint8_t * const src0 = src+7-stride;
2172 const uint8_t *src1 = src+8*stride-1;
2173 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2174 int H = src0[1] - src0[-1];
2175 int V = src1[0] - src2[ 0];
2176 for(k=2; k<=8; ++k) {
2177 src1 += stride; src2 -= stride;
2178 H += k*(src0[k] - src0[-k]);
2179 V += k*(src1[0] - src2[ 0]);
2180 }
8b82a956
MN
2181 if(svq3){
2182 H = ( 5*(H/4) ) / 16;
2183 V = ( 5*(V/4) ) / 16;
2e26c8d2
MM
2184
2185 /* required for 100% accuracy */
2186 i = H; H = V; V = i;
8b82a956
MN
2187 }else{
2188 H = ( 5*H+32 ) >> 6;
2189 V = ( 5*V+32 ) >> 6;
2190 }
30f73fc7
MN
2191
2192 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2193 for(j=16; j>0; --j) {
2194 int b = a;
2195 a += V;
2196 for(i=-16; i<0; i+=4) {
2197 src[16+i] = cm[ (b ) >> 5 ];
2198 src[17+i] = cm[ (b+ H) >> 5 ];
2199 src[18+i] = cm[ (b+2*H) >> 5 ];
2200 src[19+i] = cm[ (b+3*H) >> 5 ];
2201 b += 4*H;
0da71265 2202 }
30f73fc7
MN
2203 src += stride;
2204 }
0da71265
MN
2205}
2206
8b82a956
MN
2207static void pred16x16_plane_c(uint8_t *src, int stride){
2208 pred16x16_plane_compat_c(src, stride, 0);
2209}
2210
0da71265
MN
2211static void pred8x8_vertical_c(uint8_t *src, int stride){
2212 int i;
2213 const uint32_t a= ((uint32_t*)(src-stride))[0];
2214 const uint32_t b= ((uint32_t*)(src-stride))[1];
115329f1 2215
0da71265
MN
2216 for(i=0; i<8; i++){
2217 ((uint32_t*)(src+i*stride))[0]= a;
2218 ((uint32_t*)(src+i*stride))[1]= b;
2219 }
2220}
2221
2222static void pred8x8_horizontal_c(uint8_t *src, int stride){
2223 int i;
2224
2225 for(i=0; i<8; i++){
2226 ((uint32_t*)(src+i*stride))[0]=
2227 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2228 }
2229}
2230
2231static void pred8x8_128_dc_c(uint8_t *src, int stride){
2232 int i;
2233
43efd19a 2234 for(i=0; i<8; i++){
115329f1 2235 ((uint32_t*)(src+i*stride))[0]=
0da71265
MN
2236 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2237 }
2238}
2239
2240static void pred8x8_left_dc_c(uint8_t *src, int stride){
2241 int i;
2242 int dc0, dc2;
2243
2244 dc0=dc2=0;
2245 for(i=0;i<4; i++){
2246 dc0+= src[-1+i*stride];
2247 dc2+= src[-1+(i+4)*stride];
2248 }
2249 dc0= 0x01010101*((dc0 + 2)>>2);
2250 dc2= 0x01010101*((dc2 + 2)>>2);
2251
2252 for(i=0; i<4; i++){
2253 ((uint32_t*)(src+i*stride))[0]=
2254 ((uint32_t*)(src+i*stride))[1]= dc0;
2255 }
2256 for(i=4; i<8; i++){
2257 ((uint32_t*)(src+i*stride))[0]=
2258 ((uint32_t*)(src+i*stride))[1]= dc2;
2259 }
2260}
2261
2262static void pred8x8_top_dc_c(uint8_t *src, int stride){
2263 int i;
2264 int dc0, dc1;
2265
2266 dc0=dc1=0;
2267 for(i=0;i<4; i++){
2268 dc0+= src[i-stride];
2269 dc1+= src[4+i-stride];
2270 }
2271 dc0= 0x01010101*((dc0 + 2)>>2);
2272 dc1= 0x01010101*((dc1 + 2)>>2);
2273
2274 for(i=0; i<4; i++){
2275 ((uint32_t*)(src+i*stride))[0]= dc0;
2276 ((uint32_t*)(src+i*stride))[1]= dc1;
2277 }
2278 for(i=4; i<8; i++){
2279 ((uint32_t*)(src+i*stride))[0]= dc0;
2280 ((uint32_t*)(src+i*stride))[1]= dc1;
2281 }
2282}
2283
2284
2285static void pred8x8_dc_c(uint8_t *src, int stride){
2286 int i;
2287 int dc0, dc1, dc2, dc3;
2288
2289 dc0=dc1=dc2=0;
2290 for(i=0;i<4; i++){
2291 dc0+= src[-1+i*stride] + src[i-stride];
2292 dc1+= src[4+i-stride];
2293 dc2+= src[-1+(i+4)*stride];
2294 }
2295 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2296 dc0= 0x01010101*((dc0 + 4)>>3);
2297 dc1= 0x01010101*((dc1 + 2)>>2);
2298 dc2= 0x01010101*((dc2 + 2)>>2);
2299
2300 for(i=0; i<4; i++){
2301 ((uint32_t*)(src+i*stride))[0]= dc0;
2302 ((uint32_t*)(src+i*stride))[1]= dc1;
2303 }
2304 for(i=4; i<8; i++){
2305 ((uint32_t*)(src+i*stride))[0]= dc2;
2306 ((uint32_t*)(src+i*stride))[1]= dc3;
2307 }
2308}
2309
2310static void pred8x8_plane_c(uint8_t *src, int stride){
30f73fc7
MN
2311 int j, k;
2312 int a;
2313 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2314 const uint8_t * const src0 = src+3-stride;
2315 const uint8_t *src1 = src+4*stride-1;
2316 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2317 int H = src0[1] - src0[-1];
2318 int V = src1[0] - src2[ 0];
2319 for(k=2; k<=4; ++k) {
2320 src1 += stride; src2 -= stride;
2321 H += k*(src0[k] - src0[-k]);
2322 V += k*(src1[0] - src2[ 0]);
2323 }
2324 H = ( 17*H+16 ) >> 5;
2325 V = ( 17*V+16 ) >> 5;
2326
2327 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2328 for(j=8; j>0; --j) {
2329 int b = a;
2330 a += V;
2331 src[0] = cm[ (b ) >> 5 ];
2332 src[1] = cm[ (b+ H) >> 5 ];
2333 src[2] = cm[ (b+2*H) >> 5 ];
2334 src[3] = cm[ (b+3*H) >> 5 ];
2335 src[4] = cm[ (b+4*H) >> 5 ];
2336 src[5] = cm[ (b+5*H) >> 5 ];
2337 src[6] = cm[ (b+6*H) >> 5 ];
2338 src[7] = cm[ (b+7*H) >> 5 ];
2339 src += stride;
2340 }
0da71265
MN
2341}
2342
43efd19a
LM
2343#define SRC(x,y) src[(x)+(y)*stride]
2344#define PL(y) \
2345 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2346#define PREDICT_8x8_LOAD_LEFT \
2347 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2348 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2349 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
bf4e3bd2 2350 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
43efd19a
LM
2351
2352#define PT(x) \
2353 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2354#define PREDICT_8x8_LOAD_TOP \
2355 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2356 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2357 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
bf4e3bd2 2358 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
fb3ca6ba 2359 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
43efd19a
LM
2360
2361#define PTR(x) \
2362 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2363#define PREDICT_8x8_LOAD_TOPRIGHT \
2364 int t8, t9, t10, t11, t12, t13, t14, t15; \
2365 if(has_topright) { \
2366 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2367 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2368 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2369
2370#define PREDICT_8x8_LOAD_TOPLEFT \
fb3ca6ba 2371 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
43efd19a
LM
2372
2373#define PREDICT_8x8_DC(v) \
2374 int y; \
2375 for( y = 0; y < 8; y++ ) { \
2376 ((uint32_t*)src)[0] = \
2377 ((uint32_t*)src)[1] = v; \
2378 src += stride; \
2379 }
2380
2381static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2382{
2383 PREDICT_8x8_DC(0x80808080);
2384}
2385static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2386{
2387 PREDICT_8x8_LOAD_LEFT;
2388 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2389 PREDICT_8x8_DC(dc);
2390}
2391static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2392{
2393 PREDICT_8x8_LOAD_TOP;
2394 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2395 PREDICT_8x8_DC(dc);
2396}
2397static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2398{
2399 PREDICT_8x8_LOAD_LEFT;
2400 PREDICT_8x8_LOAD_TOP;
2401 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2402 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2403 PREDICT_8x8_DC(dc);
2404}
2405static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2406{
2407 PREDICT_8x8_LOAD_LEFT;
2408#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2409 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2410 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2411#undef ROW
2412}
2413static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2414{
2415 int y;
2416 PREDICT_8x8_LOAD_TOP;
2417 src[0] = t0;
2418 src[1] = t1;
2419 src[2] = t2;
2420 src[3] = t3;
2421 src[4] = t4;
2422 src[5] = t5;
2423 src[6] = t6;
2424 src[7] = t7;
2425 for( y = 1; y < 8; y++ )
2426 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2427}
2428static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2429{
2430 PREDICT_8x8_LOAD_TOP;
2431 PREDICT_8x8_LOAD_TOPRIGHT;
2432 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2433 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2434 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2435 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2436 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2437 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2438 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;
2439 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;
2440 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;
2441 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2442 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2443 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2444 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2445 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2446 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2447}
2448static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2449{
2450 PREDICT_8x8_LOAD_TOP;
2451 PREDICT_8x8_LOAD_LEFT;
2452 PREDICT_8x8_LOAD_TOPLEFT;
2453 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2454 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2455 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2456 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2457 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2458 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2459 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;
2460 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;
2461 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;
2462 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2463 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2464 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2465 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2466 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2467 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
115329f1 2468
43efd19a
LM
2469}
2470static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2471{
2472 PREDICT_8x8_LOAD_TOP;
2473 PREDICT_8x8_LOAD_LEFT;
2474 PREDICT_8x8_LOAD_TOPLEFT;
2475 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2476 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2477 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2478 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2479 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2480 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2481 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2482 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2483 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2484 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2485 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2486 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2487 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2488 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2489 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2490 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2491 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2492 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2493 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2494 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2495 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2496 SRC(7,0)= (t6 + t7 + 1) >> 1;
2497}
2498static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2499{
2500 PREDICT_8x8_LOAD_TOP;
2501 PREDICT_8x8_LOAD_LEFT;
2502 PREDICT_8x8_LOAD_TOPLEFT;
2503 SRC(0,7)= (l6 + l7 + 1) >> 1;
2504 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2505 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2506 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2507 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2508 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2509 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2510 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2511 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2512 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2513 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2514 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2515 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2516 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2517 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2518 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2519 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2520 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2521 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2522 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2523 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2524 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2525}
2526static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2527{
2528 PREDICT_8x8_LOAD_TOP;
2529 PREDICT_8x8_LOAD_TOPRIGHT;
2530 SRC(0,0)= (t0 + t1 + 1) >> 1;
2531 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2532 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2533 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2534 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2535 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2536 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2537 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2538 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2539 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2540 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2541 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2542 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2543 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2544 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2545 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2546 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2547 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2548 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2549 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2550 SRC(7,6)= (t10 + t11 + 1) >> 1;
2551 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2552}
2553static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2554{
2555 PREDICT_8x8_LOAD_LEFT;
2556 SRC(0,0)= (l0 + l1 + 1) >> 1;
2557 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2558 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2559 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2560 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2561 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2562 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2563 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2564 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2565 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2566 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2567 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2568 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2569 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2570 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2571 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2572 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2573 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2574}
2575#undef PREDICT_8x8_LOAD_LEFT
2576#undef PREDICT_8x8_LOAD_TOP
2577#undef PREDICT_8x8_LOAD_TOPLEFT
2578#undef PREDICT_8x8_LOAD_TOPRIGHT
2579#undef PREDICT_8x8_DC
2580#undef PTR
2581#undef PT
2582#undef PL
2583#undef SRC
2584
0da71265
MN
2585static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2586 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2587 int src_x_offset, int src_y_offset,
2588 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2589 MpegEncContext * const s = &h->s;
2590 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2591 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2592 const int luma_xy= (mx&3) + ((my&3)<<2);
2593 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2594 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2595 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2596 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2597 int extra_height= extra_width;
2598 int emu=0;
2599 const int full_mx= mx>>2;
2600 const int full_my= my>>2;
fbd312fd
LM
2601 const int pic_width = 16*s->mb_width;
2602 const int pic_height = 16*s->mb_height;
115329f1 2603
171c4076
LM
2604 if(!pic->data[0])
2605 return;
115329f1 2606
0da71265
MN
2607 if(mx&7) extra_width -= 3;
2608 if(my&7) extra_height -= 3;
115329f1
DB
2609
2610 if( full_mx < 0-extra_width
2611 || full_my < 0-extra_height
2612 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd
LM
2613 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2614 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
2615 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2616 emu=1;
2617 }
115329f1 2618
0da71265
MN
2619 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2620 if(!square){
2621 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2622 }
115329f1 2623
0da71265 2624 if(s->flags&CODEC_FLAG_GRAY) return;
115329f1 2625
0da71265 2626 if(emu){
fbd312fd 2627 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
2628 src_cb= s->edge_emu_buffer;
2629 }
2630 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2631
2632 if(emu){
fbd312fd 2633 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
2634 src_cr= s->edge_emu_buffer;
2635 }
2636 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2637}
2638
9f2d1b4f 2639static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
2640 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2641 int x_offset, int y_offset,
2642 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2643 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2644 int list0, int list1){
2645 MpegEncContext * const s = &h->s;
2646 qpel_mc_func *qpix_op= qpix_put;
2647 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 2648
0da71265
MN
2649 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2650 dest_cb += x_offset + y_offset*s->uvlinesize;
2651 dest_cr += x_offset + y_offset*s->uvlinesize;
2652 x_offset += 8*s->mb_x;
2653 y_offset += 8*s->mb_y;
115329f1 2654
0da71265 2655 if(list0){
1924f3ce 2656 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
2657 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2658 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2659 qpix_op, chroma_op);
2660
2661 qpix_op= qpix_avg;
2662 chroma_op= chroma_avg;
2663 }
2664
2665 if(list1){
1924f3ce 2666 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
2667 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2668 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2669 qpix_op, chroma_op);
2670 }
2671}
2672
9f2d1b4f
LM
2673static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2674 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2675 int x_offset, int y_offset,
2676 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2677 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2678 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2679 int list0, int list1){
2680 MpegEncContext * const s = &h->s;
2681
2682 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2683 dest_cb += x_offset + y_offset*s->uvlinesize;
2684 dest_cr += x_offset + y_offset*s->uvlinesize;
2685 x_offset += 8*s->mb_x;
2686 y_offset += 8*s->mb_y;
115329f1 2687
9f2d1b4f
LM
2688 if(list0 && list1){
2689 /* don't optimize for luma-only case, since B-frames usually
2690 * use implicit weights => chroma too. */
2691 uint8_t *tmp_cb = s->obmc_scratchpad;
2692 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2693 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2694 int refn0 = h->ref_cache[0][ scan8[n] ];
2695 int refn1 = h->ref_cache[1][ scan8[n] ];
2696
2697 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2698 dest_y, dest_cb, dest_cr,
2699 x_offset, y_offset, qpix_put, chroma_put);
2700 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2701 tmp_y, tmp_cb, tmp_cr,
2702 x_offset, y_offset, qpix_put, chroma_put);
2703
2704 if(h->use_weight == 2){
2705 int weight0 = h->implicit_weight[refn0][refn1];
2706 int weight1 = 64 - weight0;
e8b56208
LM
2707 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0);
2708 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2709 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f
LM
2710 }else{
2711 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
115329f1 2712 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
e8b56208 2713 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
9f2d1b4f 2714 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2715 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
e8b56208 2716 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
9f2d1b4f 2717 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2718 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
e8b56208 2719 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
9f2d1b4f
LM
2720 }
2721 }else{
2722 int list = list1 ? 1 : 0;
2723 int refn = h->ref_cache[list][ scan8[n] ];
2724 Picture *ref= &h->ref_list[list][refn];
2725 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2726 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2727 qpix_put, chroma_put);
2728
2729 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2730 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2731 if(h->use_weight_chroma){
2732 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2733 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2734 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2735 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2736 }
2737 }
2738}
2739
2740static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2741 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2742 int x_offset, int y_offset,
2743 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2744 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 2745 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
9f2d1b4f
LM
2746 int list0, int list1){
2747 if((h->use_weight==2 && list0 && list1
2748 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2749 || h->use_weight==1)
2750 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2751 x_offset, y_offset, qpix_put, chroma_put,
2752 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2753 else
2754 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2755 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2756}
2757
513fbd8e
LM
2758static inline void prefetch_motion(H264Context *h, int list){
2759 /* fetch pixels for estimated mv 4 macroblocks ahead
2760 * optimized for 64byte cache lines */
2761 MpegEncContext * const s = &h->s;
2762 const int refn = h->ref_cache[list][scan8[0]];
2763 if(refn >= 0){
2764 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
2765 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
2766 uint8_t **src= h->ref_list[list][refn].data;
2767 int off= mx + (my + (s->mb_x&3)*4)*s->linesize + 64;
2768 s->dsp.prefetch(src[0]+off, s->linesize, 4);
2769 off= (mx>>1) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + 64;
2770 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
2771 }
2772}
2773
0da71265
MN
2774static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2775 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
2776 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2777 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 2778 MpegEncContext * const s = &h->s;
7bc9090a 2779 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 2780 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 2781
0da71265 2782 assert(IS_INTER(mb_type));
115329f1 2783
513fbd8e
LM
2784 prefetch_motion(h, 0);
2785
0da71265
MN
2786 if(IS_16X16(mb_type)){
2787 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2788 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 2789 &weight_op[0], &weight_avg[0],
0da71265
MN
2790 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2791 }else if(IS_16X8(mb_type)){
2792 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2793 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2794 &weight_op[1], &weight_avg[1],
0da71265
MN
2795 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2796 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2797 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2798 &weight_op[1], &weight_avg[1],
0da71265
MN
2799 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2800 }else if(IS_8X16(mb_type)){
2801 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2802 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2803 &weight_op[2], &weight_avg[2],
0da71265
MN
2804 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2805 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2806 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2807 &weight_op[2], &weight_avg[2],
0da71265
MN
2808 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2809 }else{
2810 int i;
115329f1 2811
0da71265
MN
2812 assert(IS_8X8(mb_type));
2813
2814 for(i=0; i<4; i++){
2815 const int sub_mb_type= h->sub_mb_type[i];
2816 const int n= 4*i;
2817 int x_offset= (i&1)<<2;
2818 int y_offset= (i&2)<<1;
2819
2820 if(IS_SUB_8X8(sub_mb_type)){
2821 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2822 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2823 &weight_op[3], &weight_avg[3],
0da71265
MN
2824 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2825 }else if(IS_SUB_8X4(sub_mb_type)){
2826 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2827 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2828 &weight_op[4], &weight_avg[4],
0da71265
MN
2829 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2830 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2831 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2832 &weight_op[4], &weight_avg[4],
0da71265
MN
2833 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2834 }else if(IS_SUB_4X8(sub_mb_type)){
2835 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2836 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2837 &weight_op[5], &weight_avg[5],
0da71265
MN
2838 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2839 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2840 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2841 &weight_op[5], &weight_avg[5],
0da71265
MN
2842 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2843 }else{
2844 int j;
2845 assert(IS_SUB_4X4(sub_mb_type));
2846 for(j=0; j<4; j++){
2847 int sub_x_offset= x_offset + 2*(j&1);
2848 int sub_y_offset= y_offset + (j&2);
2849 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2850 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2851 &weight_op[6], &weight_avg[6],
0da71265
MN
2852 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2853 }
2854 }
2855 }
2856 }
513fbd8e
LM
2857
2858 prefetch_motion(h, 1);
0da71265
MN
2859}
2860
2861static void decode_init_vlc(H264Context *h){
2862 static int done = 0;
2863
2864 if (!done) {
2865 int i;
2866 done = 1;
2867
115329f1 2868 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
0da71265 2869 &chroma_dc_coeff_token_len [0], 1, 1,
073c2593 2870 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
0da71265
MN
2871
2872 for(i=0; i<4; i++){
115329f1 2873 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
0da71265 2874 &coeff_token_len [i][0], 1, 1,
073c2593 2875 &coeff_token_bits[i][0], 1, 1, 1);
0da71265
MN
2876 }
2877
2878 for(i=0; i<3; i++){
2879 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2880 &chroma_dc_total_zeros_len [i][0], 1, 1,
073c2593 2881 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2882 }
2883 for(i=0; i<15; i++){
115329f1 2884 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
0da71265 2885 &total_zeros_len [i][0], 1, 1,
073c2593 2886 &total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2887 }
2888
2889 for(i=0; i<6; i++){
115329f1 2890 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
0da71265 2891 &run_len [i][0], 1, 1,
073c2593 2892 &run_bits[i][0], 1, 1, 1);
0da71265 2893 }
115329f1 2894 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
0da71265 2895 &run_len [6][0], 1, 1,
073c2593 2896 &run_bits[6][0], 1, 1, 1);
0da71265
MN
2897 }
2898}
2899
2900/**
2901 * Sets the intra prediction function pointers.
2902 */
2903static void init_pred_ptrs(H264Context *h){
2904// MpegEncContext * const s = &h->s;
2905
2906 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2907 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2908 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2909 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2910 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2911 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2912 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2913 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2914 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2915 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2916 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2917 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2918
43efd19a
LM
2919 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
2920 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
2921 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
2922 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2923 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2924 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
2925 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
2926 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
2927 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
2928 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
2929 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
2930 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
2931
0da71265
MN
2932 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2933 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2934 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2935 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2936 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2937 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2938 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2939
2940 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2941 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2942 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2943 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2944 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2945 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2946 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2947}
2948
0da71265 2949static void free_tables(H264Context *h){
0da71265 2950 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
2951 av_freep(&h->chroma_pred_mode_table);
2952 av_freep(&h->cbp_table);
9e528114
LA
2953 av_freep(&h->mvd_table[0]);
2954 av_freep(&h->mvd_table[1]);
5ad984c9 2955 av_freep(&h->direct_table);
0da71265
MN
2956 av_freep(&h->non_zero_count);
2957 av_freep(&h->slice_table_base);
6ba71fc4
LLL
2958 av_freep(&h->top_borders[1]);
2959 av_freep(&h->top_borders[0]);
0da71265 2960 h->slice_table= NULL;
e5017ab8 2961
0da71265
MN
2962 av_freep(&h->mb2b_xy);
2963 av_freep(&h->mb2b8_xy);
9f2d1b4f
LM
2964
2965 av_freep(&h->s.obmc_scratchpad);
0da71265
MN
2966}
2967
239ea04c
LM
2968static void init_dequant8_coeff_table(H264Context *h){
2969 int i,q,x;
548a1c8a 2970 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
239ea04c
LM
2971 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2972 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2973
2974 for(i=0; i<2; i++ ){
2975 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2976 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2977 break;
2978 }
2979
2980 for(q=0; q<52; q++){
2981 int shift = div6[q];
2982 int idx = rem6[q];
2983 for(x=0; x<64; x++)
548a1c8a
LM
2984 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2985 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2986 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
2987 }
2988 }
2989}
2990
2991static void init_dequant4_coeff_table(H264Context *h){
2992 int i,j,q,x;
ab2e3e2c 2993 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
239ea04c
LM
2994 for(i=0; i<6; i++ ){
2995 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2996 for(j=0; j<i; j++){
2997 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2998 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2999 break;
3000 }
3001 }
3002 if(j<i)
3003 continue;
3004
3005 for(q=0; q<52; q++){
3006 int shift = div6[q] + 2;
3007 int idx = rem6[q];
3008 for(x=0; x<16; x++)
ab2e3e2c
LM
3009 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
3010 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
3011 h->pps.scaling_matrix4[i][x]) << shift;
3012 }
3013 }
3014}
3015
3016static void init_dequant_tables(H264Context *h){
3017 int i,x;
3018 init_dequant4_coeff_table(h);
3019 if(h->pps.transform_8x8_mode)
3020 init_dequant8_coeff_table(h);
3021 if(h->sps.transform_bypass){
3022 for(i=0; i<6; i++)
3023 for(x=0; x<16; x++)
3024 h->dequant4_coeff[i][0][x] = 1<<6;
3025 if(h->pps.transform_8x8_mode)
3026 for(i=0; i<2; i++)
3027 for(x=0; x<64; x++)
3028 h->dequant8_coeff[i][0][x] = 1<<6;
3029 }
3030}
3031
3032
0da71265
MN
3033/**
3034 * allocates tables.
3b66c4c5 3035 * needs width/height
0da71265
MN
3036 */
3037static int alloc_tables(H264Context *h){
3038 MpegEncContext * const s = &h->s;
7bc9090a 3039 const int big_mb_num= s->mb_stride * (s->mb_height+1);
239ea04c 3040 int x,y;
0da71265
MN
3041
3042 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 3043
53c05b1e 3044 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
0da71265 3045 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
6ba71fc4
LLL
3046 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
3047 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
5d0e4cb8 3048 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 3049
e5017ab8
LA
3050 if( h->pps.cabac ) {
3051 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
9e528114
LA
3052 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3053 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
5ad984c9 3054 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8
LA
3055 }
3056
0da71265 3057 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
7bc9090a 3058 h->slice_table= h->slice_table_base + s->mb_stride + 1;
0da71265 3059
a55f20bd
LM
3060 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
3061 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
0da71265
MN
3062 for(y=0; y<s->mb_height; y++){
3063 for(x=0; x<s->mb_width; x++){
7bc9090a 3064 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
3065 const int b_xy = 4*x + 4*y*h->b_stride;
3066 const int b8_xy= 2*x + 2*y*h->b8_stride;
115329f1 3067
0da71265
MN
3068 h->mb2b_xy [mb_xy]= b_xy;
3069 h->mb2b8_xy[mb_xy]= b8_xy;
3070 }
3071 }
9f2d1b4f 3072
9c6221ae
GV
3073 s->obmc_scratchpad = NULL;
3074
56edbd81
LM
3075 if(!h->dequant4_coeff[0])
3076 init_dequant_tables(h);
3077
0da71265
MN
3078 return 0;
3079fail:
3080 free_tables(h);
3081 return -1;
3082}
3083
3084static void common_init(H264Context *h){
3085 MpegEncContext * const s = &h->s;
0da71265
MN
3086
3087 s->width = s->avctx->width;
3088 s->height = s->avctx->height;
3089 s->codec_id= s->avctx->codec->id;
115329f1 3090
0da71265
MN
3091 init_pred_ptrs(h);
3092
239ea04c 3093 h->dequant_coeff_pps= -1;
9a41c2c7 3094 s->unrestricted_mv=1;
0da71265 3095 s->decode=1; //FIXME
56edbd81
LM
3096
3097 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3098 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
3099}
3100
3101static int decode_init(AVCodecContext *avctx){
3102 H264Context *h= avctx->priv_data;
3103 MpegEncContext * const s = &h->s;
3104
3edcacde 3105 MPV_decode_defaults(s);
115329f1 3106
0da71265
MN
3107 s->avctx = avctx;
3108 common_init(h);
3109
3110 s->out_format = FMT_H264;
3111 s->workaround_bugs= avctx->workaround_bugs;
3112
3113 // set defaults
0da71265
MN
3114// s->decode_mb= ff_h263_decode_mb;
3115 s->low_delay= 1;
3116 avctx->pix_fmt= PIX_FMT_YUV420P;
3117
3118 decode_init_vlc(h);
115329f1 3119
26165f99
MR
3120 if(avctx->extradata_size > 0 && avctx->extradata &&
3121 *(char *)avctx->extradata == 1){
4770b1b4
RT
3122 h->is_avc = 1;
3123 h->got_avcC = 0;
26165f99
MR
3124 } else {
3125 h->is_avc = 0;
4770b1b4
RT
3126 }
3127
0da71265
MN
3128 return 0;
3129}
3130
af8aa846 3131static int frame_start(H264Context *h){
0da71265
MN
3132 MpegEncContext * const s = &h->s;
3133 int i;
3134
af8aa846
MN
3135 if(MPV_frame_start(s, s->avctx) < 0)
3136 return -1;
0da71265 3137 ff_er_frame_start(s);
0da71265
MN
3138
3139 assert(s->linesize && s->uvlinesize);
3140
3141 for(i=0; i<16; i++){
3142 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 3143 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3144 }
3145 for(i=0; i<4; i++){
3146 h->block_offset[16+i]=
3147 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
3148 h->block_offset[24+16+i]=
3149 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3150 }
3151
934b0821
LM
3152 /* can't be in alloc_tables because linesize isn't known there.
3153 * FIXME: redo bipred weight to not require extra buffer? */
3154 if(!s->obmc_scratchpad)
3155 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3156
0da71265 3157// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
af8aa846 3158 return 0;
0da71265
MN
3159}
3160
53c05b1e
MN
3161static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3162 MpegEncContext * const s = &h->s;
3163 int i;
115329f1 3164
53c05b1e
MN
3165 src_y -= linesize;
3166 src_cb -= uvlinesize;
3167 src_cr -= uvlinesize;
3168
3b66c4c5 3169 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 3170 // and the line above the bottom macroblock
6ba71fc4 3171 h->left_border[0]= h->top_borders[0][s->mb_x][15];
53c05b1e
MN
3172 for(i=1; i<17; i++){
3173 h->left_border[i]= src_y[15+i* linesize];
3174 }
115329f1 3175
6ba71fc4
LLL
3176 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3177 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
53c05b1e
MN
3178
3179 if(!(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
3180 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3181 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
53c05b1e
MN
3182 for(i=1; i<9; i++){
3183 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3184 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3185 }
6ba71fc4
LLL
3186 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3187 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
53c05b1e
MN
3188 }
3189}
3190
3191static 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){
3192 MpegEncContext * const s = &h->s;
3193 int temp8, i;
3194 uint64_t temp64;
d89dc06a
LM
3195 int deblock_left = (s->mb_x > 0);
3196 int deblock_top = (s->mb_y > 0);
53c05b1e
MN
3197
3198 src_y -= linesize + 1;
3199 src_cb -= uvlinesize + 1;
3200 src_cr -= uvlinesize + 1;
3201
3202#define XCHG(a,b,t,xchg)\
3203t= a;\
3204if(xchg)\
3205 a= b;\
3206b= t;
d89dc06a
LM
3207
3208 if(deblock_left){
3209 for(i = !deblock_top; i<17; i++){
3210 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3211 }
3212 }
3213
3214 if(deblock_top){
6ba71fc4
LLL
3215 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3216 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
cad4368a 3217 if(s->mb_x+1 < s->mb_width){
43efd19a
LM
3218 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3219 }
53c05b1e 3220 }
53c05b1e
MN
3221
3222 if(!(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
3223 if(deblock_left){
3224 for(i = !deblock_top; i<9; i++){
3225 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3226 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3227 }
3228 }
3229 if(deblock_top){
6ba71fc4
LLL
3230 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3231 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3232 }
3233 }
3234}
3235
3236static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3237 MpegEncContext * const s = &h->s;
3238 int i;
115329f1 3239
6ba71fc4
LLL
3240 src_y -= 2 * linesize;
3241 src_cb -= 2 * uvlinesize;
3242 src_cr -= 2 * uvlinesize;
3243
3b66c4c5 3244 // There are two lines saved, the line above the the top macroblock of a pair,
6ba71fc4
LLL
3245 // and the line above the bottom macroblock
3246 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3247 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3248 for(i=2; i<34; i++){
3249 h->left_border[i]= src_y[15+i* linesize];
3250 }
115329f1 3251
6ba71fc4
LLL
3252 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3253 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3254 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3255 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3256
3257 if(!(s->flags&CODEC_FLAG_GRAY)){
3258 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3259 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3260 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3261 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3262 for(i=2; i<18; i++){
3263 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3264 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3265 }
3266 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3267 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3268 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3269 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3270 }
3271}
3272
3273static 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){
3274 MpegEncContext * const s = &h->s;
3275 int temp8, i;
3276 uint64_t temp64;
3277 int deblock_left = (s->mb_x > 0);
3278 int deblock_top = (s->mb_y > 0);
3279
3280 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);
3281
3282 src_y -= 2 * linesize + 1;
3283 src_cb -= 2 * uvlinesize + 1;
3284 src_cr -= 2 * uvlinesize + 1;
3285
3286#define XCHG(a,b,t,xchg)\
3287t= a;\
3288if(xchg)\
3289 a= b;\
3290b= t;
3291
3292 if(deblock_left){
3293 for(i = (!deblock_top)<<1; i<34; i++){
3294 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3295 }
3296 }
3297
3298 if(deblock_top){
3299 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3300 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3301 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3302 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3303 }
3304
3305 if(!(s->flags&CODEC_FLAG_GRAY)){
3306 if(deblock_left){
3307 for(i = (!deblock_top) << 1; i<18; i++){
3308 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
3309 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3310 }
3311 }
3312 if(deblock_top){
3313 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3314 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3315 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3316 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
53c05b1e 3317 }
53c05b1e
MN
3318 }
3319}
3320
0da71265
MN
3321static void hl_decode_mb(H264Context *h){
3322 MpegEncContext * const s = &h->s;
3323 const int mb_x= s->mb_x;
3324 const int mb_y= s->mb_y;
7bc9090a 3325 const int mb_xy= mb_x + mb_y*s->mb_stride;
0da71265
MN
3326 const int mb_type= s->current_picture.mb_type[mb_xy];
3327 uint8_t *dest_y, *dest_cb, *dest_cr;
3328 int linesize, uvlinesize /*dct_offset*/;
3329 int i;
6867a90b 3330 int *block_offset = &h->block_offset[0];
6ba71fc4 3331 const unsigned int bottom = mb_y & 1;
36940eca
LM
3332 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3333 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15 3334 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
0da71265
MN
3335
3336 if(!s->decode)
3337 return;
3338
0da71265
MN
3339 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3340 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3341 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3342
3343 if (h->mb_field_decoding_flag) {
3344 linesize = s->linesize * 2;
3345 uvlinesize = s->uvlinesize * 2;
6867a90b 3346 block_offset = &h->block_offset[24];
0da71265
MN
3347 if(mb_y&1){ //FIXME move out of this func?
3348 dest_y -= s->linesize*15;
6867a90b
LLL
3349 dest_cb-= s->uvlinesize*7;
3350 dest_cr-= s->uvlinesize*7;
0da71265
MN
3351 }
3352 } else {
3353 linesize = s->linesize;
3354 uvlinesize = s->uvlinesize;
3355// dct_offset = s->linesize * 16;
3356 }
115329f1 3357
ef9d1d15
LM
3358 if(transform_bypass){
3359 idct_dc_add =
3360 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3361 }else if(IS_8x8DCT(mb_type)){
3362 idct_dc_add = s->dsp.h264_idct8_dc_add;
3363 idct_add = s->dsp.h264_idct8_add;
3364 }else{
3365 idct_dc_add = s->dsp.h264_idct_dc_add;
3366 idct_add = s->dsp.h264_idct_add;
3367 }
0da71265 3368
6fbcaaa0
LLL
3369 if (IS_INTRA_PCM(mb_type)) {
3370 unsigned int x, y;
3371
3372 // The pixels are stored in h->mb array in the same order as levels,
3373 // copy them in output in the correct order.
3374 for(i=0; i<16; i++) {
3375 for (y=0; y<4; y++) {
3376 for (x=0; x<4; x++) {
6867a90b 3377 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3378 }
3379 }
3380 }
3381 for(i=16; i<16+4; i++) {
3382 for (y=0; y<4; y++) {
3383 for (x=0; x<4; x++) {
6867a90b 3384 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3385 }
3386 }
3387 }
3388 for(i=20; i<20+4; i++) {
3389 for (y=0; y<4; y++) {
3390 for (x=0; x<4; x++) {
6867a90b 3391 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3392 }
3393 }
3394 }
e7e09b49
LLL
3395 } else {
3396 if(IS_INTRA(mb_type)){
6ba71fc4
LLL
3397 if(h->deblocking_filter) {
3398 if (h->mb_aff_frame) {
3399 if (!bottom)
3400 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3401 } else {
3402 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3403 }
3404 }
53c05b1e 3405
e7e09b49
LLL
3406 if(!(s->flags&CODEC_FLAG_GRAY)){
3407 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3408 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3409 }
0da71265 3410
e7e09b49
LLL
3411 if(IS_INTRA4x4(mb_type)){
3412 if(!s->encoding){
43efd19a
LM
3413 if(IS_8x8DCT(mb_type)){
3414 for(i=0; i<16; i+=4){
3415 uint8_t * const ptr= dest_y + block_offset[i];
3416 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3417 const int nnz = h->non_zero_count_cache[ scan8[i] ];
43efd19a
LM
3418 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3419 (h->topright_samples_available<<(i+1))&0x8000, linesize);
ef9d1d15
LM
3420 if(nnz){
3421 if(nnz == 1 && h->mb[i*16])
3422 idct_dc_add(ptr, h->mb + i*16, linesize);
3423 else
3424 idct_add(ptr, h->mb + i*16, linesize);
3425 }
43efd19a
LM
3426 }
3427 }else
e7e09b49 3428 for(i=0; i<16; i++){
6867a90b 3429 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
3430 uint8_t *topright;
3431 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3432 int nnz, tr;
e7e09b49
LLL
3433
3434 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3435 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6867a90b 3436 assert(mb_y || linesize <= block_offset[i]);
e7e09b49
LLL
3437 if(!topright_avail){
3438 tr= ptr[3 - linesize]*0x01010101;
3439 topright= (uint8_t*) &tr;
115329f1 3440 }else
e7e09b49 3441 topright= ptr + 4 - linesize;
a9799653 3442 }else
e7e09b49
LLL
3443 topright= NULL;
3444
3445 h->pred4x4[ dir ](ptr, topright, linesize);
ef9d1d15
LM
3446 nnz = h->non_zero_count_cache[ scan8[i] ];
3447 if(nnz){
3448 if(s->codec_id == CODEC_ID_H264){
3449 if(nnz == 1 && h->mb[i*16])
3450 idct_dc_add(ptr, h->mb + i*16, linesize);
3451 else
3452 idct_add(ptr, h->mb + i*16, linesize);
3453 }else
e7e09b49
LLL
3454 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3455 }
8b82a956 3456 }
0da71265 3457 }
e7e09b49
LLL
3458 }else{
3459 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
36940eca
LM
3460 if(s->codec_id == CODEC_ID_H264){
3461 if(!transform_bypass)
239ea04c 3462 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
36940eca 3463 }else
e7e09b49 3464 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
0da71265 3465 }
6ba71fc4
LLL
3466 if(h->deblocking_filter) {
3467 if (h->mb_aff_frame) {
3468 if (bottom) {
3469 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3470 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3471 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3472 s->mb_y--;
3473 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3474 s->mb_y++;
3475 }
3476 } else {
3477 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3478 }
3479 }
e7e09b49
LLL
3480 }else if(s->codec_id == CODEC_ID_H264){
3481 hl_motion(h, dest_y, dest_cb, dest_cr,
115329f1 3482 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
e7e09b49
LLL
3483 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3484 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
0da71265 3485 }
e7e09b49
LLL
3486
3487
3488 if(!IS_INTRA4x4(mb_type)){
3489 if(s->codec_id == CODEC_ID_H264){
ef9d1d15
LM
3490 if(IS_INTRA16x16(mb_type)){
3491 for(i=0; i<16; i++){
3492 if(h->non_zero_count_cache[ scan8[i] ])
3493 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3494 else if(h->mb[i*16])
3495 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3496 }
3497 }else{
3498 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3499 for(i=0; i<16; i+=di){
3500 int nnz = h->non_zero_count_cache[ scan8[i] ];