use 64bit types
[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
0da71265
MN
36#undef NDEBUG
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
1425 for(list=0; list<2; list++){
1426 int y;
53b19144 1427 if(!USES_LIST(mb_type, list))
5ad984c9 1428 continue;
115329f1 1429
0da71265
MN
1430 for(y=0; y<4; y++){
1431 *(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];
1432 *(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];
1433 }
9e528114
LA
1434 if( h->pps.cabac ) {
1435 for(y=0; y<4; y++){
1436 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1437 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1438 }
1439 }
53b19144
LM
1440
1441 {
1442 uint8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1443 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1444 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1445 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1446 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
0da71265
MN
1447 }
1448 }
115329f1 1449
5ad984c9
LM
1450 if(h->slice_type == B_TYPE && h->pps.cabac){
1451 if(IS_8X8(mb_type)){
53b19144
LM
1452 uint8_t *direct_table = &h->direct_table[b8_xy];
1453 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1454 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1455 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
5ad984c9
LM
1456 }
1457 }
0da71265
MN
1458}
1459
1460/**
1461 * Decodes a network abstraction layer unit.
1462 * @param consumed is the number of bytes used as input
1463 * @param length is the length of the array
3b66c4c5 1464 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
115329f1 1465 * @returns decoded bytes, might be src+1 if no escapes
0da71265
MN
1466 */
1467static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1468 int i, si, di;
1469 uint8_t *dst;
1470
bb270c08 1471// src[0]&0x80; //forbidden bit
0da71265
MN
1472 h->nal_ref_idc= src[0]>>5;
1473 h->nal_unit_type= src[0]&0x1F;
1474
1475 src++; length--;
115329f1 1476#if 0
0da71265
MN
1477 for(i=0; i<length; i++)
1478 printf("%2X ", src[i]);
1479#endif
1480 for(i=0; i+1<length; i+=2){
1481 if(src[i]) continue;
1482 if(i>0 && src[i-1]==0) i--;
1483 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1484 if(src[i+2]!=3){
1485 /* startcode, so we must be past the end */
1486 length=i;
1487 }
1488 break;
1489 }
1490 }
1491
1492 if(i>=length-1){ //no escaped 0
1493 *dst_length= length;
1494 *consumed= length+1; //+1 for the header
115329f1 1495 return src;
0da71265
MN
1496 }
1497
1498 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1499 dst= h->rbsp_buffer;
1500
3b66c4c5 1501//printf("decoding esc\n");
0da71265 1502 si=di=0;
115329f1 1503 while(si<length){
0da71265
MN
1504 //remove escapes (very rare 1:2^22)
1505 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1506 if(src[si+2]==3){ //escape
1507 dst[di++]= 0;
1508 dst[di++]= 0;
1509 si+=3;
c8470cc1 1510 continue;
0da71265
MN
1511 }else //next start code
1512 break;
1513 }
1514
1515 dst[di++]= src[si++];
1516 }
1517
1518 *dst_length= di;
1519 *consumed= si + 1;//+1 for the header
1520//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1521 return dst;
1522}
1523
e5017ab8 1524#if 0
0da71265
MN
1525/**
1526 * @param src the data which should be escaped
1527 * @param dst the target buffer, dst+1 == src is allowed as a special case
1528 * @param length the length of the src data
1529 * @param dst_length the length of the dst array
1530 * @returns length of escaped data in bytes or -1 if an error occured
1531 */
1532static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1533 int i, escape_count, si, di;
1534 uint8_t *temp;
115329f1 1535
0da71265
MN
1536 assert(length>=0);
1537 assert(dst_length>0);
115329f1 1538
0da71265
MN
1539 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1540
1541 if(length==0) return 1;
1542
1543 escape_count= 0;
1544 for(i=0; i<length; i+=2){
1545 if(src[i]) continue;
115329f1 1546 if(i>0 && src[i-1]==0)
0da71265
MN
1547 i--;
1548 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1549 escape_count++;
1550 i+=2;
1551 }
1552 }
115329f1
DB
1553
1554 if(escape_count==0){
0da71265
MN
1555 if(dst+1 != src)
1556 memcpy(dst+1, src, length);
1557 return length + 1;
1558 }
115329f1 1559
0da71265
MN
1560 if(length + escape_count + 1> dst_length)
1561 return -1;
1562
1563 //this should be damn rare (hopefully)
1564
1565 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1566 temp= h->rbsp_buffer;
1567//printf("encoding esc\n");
115329f1 1568
0da71265
MN
1569 si= 0;
1570 di= 0;
1571 while(si < length){
1572 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1573 temp[di++]= 0; si++;
1574 temp[di++]= 0; si++;
115329f1 1575 temp[di++]= 3;
0da71265
MN
1576 temp[di++]= src[si++];
1577 }
1578 else
1579 temp[di++]= src[si++];
1580 }
1581 memcpy(dst+1, temp, length+escape_count);
115329f1 1582
0da71265 1583 assert(di == length+escape_count);
115329f1 1584
0da71265
MN
1585 return di + 1;
1586}
1587
1588/**
1589 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1590 */
1591static void encode_rbsp_trailing(PutBitContext *pb){
1592 int length;
1593 put_bits(pb, 1, 1);
fe455f33 1594 length= (-put_bits_count(pb))&7;
0da71265
MN
1595 if(length) put_bits(pb, length, 0);
1596}
e5017ab8 1597#endif
0da71265
MN
1598
1599/**
1600 * identifies the exact end of the bitstream
1601 * @return the length of the trailing, or 0 if damaged
1602 */
1603static int decode_rbsp_trailing(uint8_t *src){
1604 int v= *src;
1605 int r;
1606
95c26348 1607 tprintf("rbsp trailing %X\n", v);
0da71265
MN
1608
1609 for(r=1; r<9; r++){
1610 if(v&1) return r;
1611 v>>=1;
1612 }
1613 return 0;
1614}
1615
1616/**
1617 * idct tranforms the 16 dc values and dequantize them.
1618 * @param qp quantization parameter
1619 */
239ea04c 1620static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1621#define stride 16
1622 int i;
1623 int temp[16]; //FIXME check if this is a good idea
1624 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1625 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1626
1627//memset(block, 64, 2*256);
1628//return;
1629 for(i=0; i<4; i++){
1630 const int offset= y_offset[i];
1631 const int z0= block[offset+stride*0] + block[offset+stride*4];
1632 const int z1= block[offset+stride*0] - block[offset+stride*4];
1633 const int z2= block[offset+stride*1] - block[offset+stride*5];
1634 const int z3= block[offset+stride*1] + block[offset+stride*5];
1635
1636 temp[4*i+0]= z0+z3;
1637 temp[4*i+1]= z1+z2;
1638 temp[4*i+2]= z1-z2;
1639 temp[4*i+3]= z0-z3;
1640 }
1641
1642 for(i=0; i<4; i++){
1643 const int offset= x_offset[i];
1644 const int z0= temp[4*0+i] + temp[4*2+i];
1645 const int z1= temp[4*0+i] - temp[4*2+i];
1646 const int z2= temp[4*1+i] - temp[4*3+i];
1647 const int z3= temp[4*1+i] + temp[4*3+i];
1648
239ea04c
LM
1649 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1650 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1651 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1652 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
0da71265
MN
1653 }
1654}
1655
e5017ab8 1656#if 0
0da71265
MN
1657/**
1658 * dct tranforms the 16 dc values.
1659 * @param qp quantization parameter ??? FIXME
1660 */
1661static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1662// const int qmul= dequant_coeff[qp][0];
1663 int i;
1664 int temp[16]; //FIXME check if this is a good idea
1665 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1666 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1667
1668 for(i=0; i<4; i++){
1669 const int offset= y_offset[i];
1670 const int z0= block[offset+stride*0] + block[offset+stride*4];
1671 const int z1= block[offset+stride*0] - block[offset+stride*4];
1672 const int z2= block[offset+stride*1] - block[offset+stride*5];
1673 const int z3= block[offset+stride*1] + block[offset+stride*5];
1674
1675 temp[4*i+0]= z0+z3;
1676 temp[4*i+1]= z1+z2;
1677 temp[4*i+2]= z1-z2;
1678 temp[4*i+3]= z0-z3;
1679 }
1680
1681 for(i=0; i<4; i++){
1682 const int offset= x_offset[i];
1683 const int z0= temp[4*0+i] + temp[4*2+i];
1684 const int z1= temp[4*0+i] - temp[4*2+i];
1685 const int z2= temp[4*1+i] - temp[4*3+i];
1686 const int z3= temp[4*1+i] + temp[4*3+i];
1687
1688 block[stride*0 +offset]= (z0 + z3)>>1;
1689 block[stride*2 +offset]= (z1 + z2)>>1;
1690 block[stride*8 +offset]= (z1 - z2)>>1;
1691 block[stride*10+offset]= (z0 - z3)>>1;
1692 }
1693}
e5017ab8
LA
1694#endif
1695
0da71265
MN
1696#undef xStride
1697#undef stride
1698
239ea04c 1699static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
0da71265
MN
1700 const int stride= 16*2;
1701 const int xStride= 16;
1702 int a,b,c,d,e;
1703
1704 a= block[stride*0 + xStride*0];
1705 b= block[stride*0 + xStride*1];
1706 c= block[stride*1 + xStride*0];
1707 d= block[stride*1 + xStride*1];
1708
1709 e= a-b;
1710 a= a+b;
1711 b= c-d;
1712 c= c+d;
1713
239ea04c
LM
1714 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1715 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1716 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1717 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
0da71265
MN
1718}
1719
e5017ab8 1720#if 0
0da71265
MN
1721static void chroma_dc_dct_c(DCTELEM *block){
1722 const int stride= 16*2;
1723 const int xStride= 16;
1724 int a,b,c,d,e;
1725
1726 a= block[stride*0 + xStride*0];
1727 b= block[stride*0 + xStride*1];
1728 c= block[stride*1 + xStride*0];
1729 d= block[stride*1 + xStride*1];
1730
1731 e= a-b;
1732 a= a+b;
1733 b= c-d;
1734 c= c+d;
1735
1736 block[stride*0 + xStride*0]= (a+c);
1737 block[stride*0 + xStride*1]= (e+b);
1738 block[stride*1 + xStride*0]= (a-c);
1739 block[stride*1 + xStride*1]= (e-b);
1740}
e5017ab8 1741#endif
0da71265
MN
1742
1743/**
1744 * gets the chroma qp.
1745 */
6ba71fc4 1746static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
115329f1 1747
6ba71fc4 1748 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
0da71265
MN
1749}
1750
1751
e5017ab8 1752#if 0
0da71265
MN
1753static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1754 int i;
1755 //FIXME try int temp instead of block
115329f1 1756
0da71265
MN
1757 for(i=0; i<4; i++){
1758 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1759 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1760 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1761 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1762 const int z0= d0 + d3;
1763 const int z3= d0 - d3;
1764 const int z1= d1 + d2;
1765 const int z2= d1 - d2;
115329f1 1766
0da71265
MN
1767 block[0 + 4*i]= z0 + z1;
1768 block[1 + 4*i]= 2*z3 + z2;
1769 block[2 + 4*i]= z0 - z1;
1770 block[3 + 4*i]= z3 - 2*z2;
115329f1 1771 }
0da71265
MN
1772
1773 for(i=0; i<4; i++){
1774 const int z0= block[0*4 + i] + block[3*4 + i];
1775 const int z3= block[0*4 + i] - block[3*4 + i];
1776 const int z1= block[1*4 + i] + block[2*4 + i];
1777 const int z2= block[1*4 + i] - block[2*4 + i];
115329f1 1778
0da71265
MN
1779 block[0*4 + i]= z0 + z1;
1780 block[1*4 + i]= 2*z3 + z2;
1781 block[2*4 + i]= z0 - z1;
1782 block[3*4 + i]= z3 - 2*z2;
1783 }
1784}
e5017ab8 1785#endif
0da71265 1786
3b66c4c5 1787//FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
0da71265
MN
1788//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1789static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1790 int i;
1791 const int * const quant_table= quant_coeff[qscale];
1792 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1793 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1794 const unsigned int threshold2= (threshold1<<1);
1795 int last_non_zero;
1796
1797 if(seperate_dc){
1798 if(qscale<=18){
1799 //avoid overflows
1800 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1801 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1802 const unsigned int dc_threshold2= (dc_threshold1<<1);
1803
1804 int level= block[0]*quant_coeff[qscale+18][0];
1805 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1806 if(level>0){
1807 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1808 block[0]= level;
1809 }else{
1810 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1811 block[0]= -level;
1812 }
1813// last_non_zero = i;
1814 }else{
1815 block[0]=0;
1816 }
1817 }else{
1818 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1819 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1820 const unsigned int dc_threshold2= (dc_threshold1<<1);
1821
1822 int level= block[0]*quant_table[0];
1823 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1824 if(level>0){
1825 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1826 block[0]= level;
1827 }else{
1828 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1829 block[0]= -level;
1830 }
1831// last_non_zero = i;
1832 }else{
1833 block[0]=0;
1834 }
1835 }
1836 last_non_zero= 0;
1837 i=1;
1838 }else{
1839 last_non_zero= -1;
1840 i=0;
1841 }
1842
1843 for(; i<16; i++){
1844 const int j= scantable[i];
1845 int level= block[j]*quant_table[j];
1846
1847// if( bias+level >= (1<<(QMAT_SHIFT - 3))
1848// || bias-level >= (1<<(QMAT_SHIFT - 3))){
1849 if(((unsigned)(level+threshold1))>threshold2){
1850 if(level>0){
1851 level= (bias + level)>>QUANT_SHIFT;
1852 block[j]= level;
1853 }else{
1854 level= (bias - level)>>QUANT_SHIFT;
1855 block[j]= -level;
1856 }
1857 last_non_zero = i;
1858 }else{
1859 block[j]=0;
1860 }
1861 }
1862
1863 return last_non_zero;
1864}
1865
1866static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1867 const uint32_t a= ((uint32_t*)(src-stride))[0];
1868 ((uint32_t*)(src+0*stride))[0]= a;
1869 ((uint32_t*)(src+1*stride))[0]= a;
1870 ((uint32_t*)(src+2*stride))[0]= a;
1871 ((uint32_t*)(src+3*stride))[0]= a;
1872}
1873
1874static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1875 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1876 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1877 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1878 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1879}
1880
1881static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1882 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1883 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
115329f1
DB
1884
1885 ((uint32_t*)(src+0*stride))[0]=
1886 ((uint32_t*)(src+1*stride))[0]=
1887 ((uint32_t*)(src+2*stride))[0]=
1888 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1889}
1890
1891static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1892 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
115329f1
DB
1893
1894 ((uint32_t*)(src+0*stride))[0]=
1895 ((uint32_t*)(src+1*stride))[0]=
1896 ((uint32_t*)(src+2*stride))[0]=
1897 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1898}
1899
1900static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1901 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
115329f1
DB
1902
1903 ((uint32_t*)(src+0*stride))[0]=
1904 ((uint32_t*)(src+1*stride))[0]=
1905 ((uint32_t*)(src+2*stride))[0]=
1906 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
0da71265
MN
1907}
1908
1909static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1910 ((uint32_t*)(src+0*stride))[0]=
1911 ((uint32_t*)(src+1*stride))[0]=
1912 ((uint32_t*)(src+2*stride))[0]=
0da71265
MN
1913 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1914}
1915
1916
1917#define LOAD_TOP_RIGHT_EDGE\
1918 const int t4= topright[0];\
1919 const int t5= topright[1];\
1920 const int t6= topright[2];\
1921 const int t7= topright[3];\
1922
1923#define LOAD_LEFT_EDGE\
1924 const int l0= src[-1+0*stride];\
1925 const int l1= src[-1+1*stride];\
1926 const int l2= src[-1+2*stride];\
1927 const int l3= src[-1+3*stride];\
1928
1929#define LOAD_TOP_EDGE\
1930 const int t0= src[ 0-1*stride];\
1931 const int t1= src[ 1-1*stride];\
1932 const int t2= src[ 2-1*stride];\
1933 const int t3= src[ 3-1*stride];\
1934
1935static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1936 const int lt= src[-1-1*stride];
1937 LOAD_TOP_EDGE
1938 LOAD_LEFT_EDGE
1939
115329f1 1940 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
0da71265 1941 src[0+2*stride]=
115329f1 1942 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
0da71265
MN
1943 src[0+1*stride]=
1944 src[1+2*stride]=
115329f1 1945 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
0da71265
MN
1946 src[0+0*stride]=
1947 src[1+1*stride]=
1948 src[2+2*stride]=
115329f1 1949 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
0da71265
MN
1950 src[1+0*stride]=
1951 src[2+1*stride]=
1952 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1953 src[2+0*stride]=
1954 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1955 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
4cfbf61b 1956}
0da71265
MN
1957
1958static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
1959 LOAD_TOP_EDGE
1960 LOAD_TOP_RIGHT_EDGE
1961// LOAD_LEFT_EDGE
0da71265
MN
1962
1963 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1964 src[1+0*stride]=
1965 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1966 src[2+0*stride]=
1967 src[1+1*stride]=
1968 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1969 src[3+0*stride]=
1970 src[2+1*stride]=
1971 src[1+2*stride]=
1972 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1973 src[3+1*stride]=
1974 src[2+2*stride]=
1975 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1976 src[3+2*stride]=
1977 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1978 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
4cfbf61b 1979}
0da71265
MN
1980
1981static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1982 const int lt= src[-1-1*stride];
115329f1
DB
1983 LOAD_TOP_EDGE
1984 LOAD_LEFT_EDGE
0da71265
MN
1985 const __attribute__((unused)) int unu= l3;
1986
1987 src[0+0*stride]=
1988 src[1+2*stride]=(lt + t0 + 1)>>1;
1989 src[1+0*stride]=
1990 src[2+2*stride]=(t0 + t1 + 1)>>1;
1991 src[2+0*stride]=
1992 src[3+2*stride]=(t1 + t2 + 1)>>1;
1993 src[3+0*stride]=(t2 + t3 + 1)>>1;
1994 src[0+1*stride]=
1995 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1996 src[1+1*stride]=
1997 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1998 src[2+1*stride]=
1999 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2000 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2001 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2002 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
4cfbf61b 2003}
0da71265
MN
2004
2005static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
115329f1
DB
2006 LOAD_TOP_EDGE
2007 LOAD_TOP_RIGHT_EDGE
0da71265
MN
2008 const __attribute__((unused)) int unu= t7;
2009
2010 src[0+0*stride]=(t0 + t1 + 1)>>1;
2011 src[1+0*stride]=
2012 src[0+2*stride]=(t1 + t2 + 1)>>1;
2013 src[2+0*stride]=
2014 src[1+2*stride]=(t2 + t3 + 1)>>1;
2015 src[3+0*stride]=
2016 src[2+2*stride]=(t3 + t4+ 1)>>1;
2017 src[3+2*stride]=(t4 + t5+ 1)>>1;
2018 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2019 src[1+1*stride]=
2020 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2021 src[2+1*stride]=
2022 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2023 src[3+1*stride]=
2024 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2025 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
4cfbf61b 2026}
0da71265
MN
2027
2028static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
115329f1 2029 LOAD_LEFT_EDGE
0da71265
MN
2030
2031 src[0+0*stride]=(l0 + l1 + 1)>>1;
2032 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2033 src[2+0*stride]=
2034 src[0+1*stride]=(l1 + l2 + 1)>>1;
2035 src[3+0*stride]=
2036 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2037 src[2+1*stride]=
2038 src[0+2*stride]=(l2 + l3 + 1)>>1;
2039 src[3+1*stride]=
2040 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2041 src[3+2*stride]=
2042 src[1+3*stride]=
2043 src[0+3*stride]=
2044 src[2+2*stride]=
2045 src[2+3*stride]=
2046 src[3+3*stride]=l3;
4cfbf61b 2047}
115329f1 2048
0da71265
MN
2049static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2050 const int lt= src[-1-1*stride];
115329f1
DB
2051 LOAD_TOP_EDGE
2052 LOAD_LEFT_EDGE
0da71265
MN
2053 const __attribute__((unused)) int unu= t3;
2054
2055 src[0+0*stride]=
2056 src[2+1*stride]=(lt + l0 + 1)>>1;
2057 src[1+0*stride]=
2058 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2059 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2060 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2061 src[0+1*stride]=
2062 src[2+2*stride]=(l0 + l1 + 1)>>1;
2063 src[1+1*stride]=
2064 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2065 src[0+2*stride]=
2066 src[2+3*stride]=(l1 + l2+ 1)>>1;
2067 src[1+2*stride]=
2068 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2069 src[0+3*stride]=(l2 + l3 + 1)>>1;
2070 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
4cfbf61b 2071}
0da71265
MN
2072
2073static void pred16x16_vertical_c(uint8_t *src, int stride){
2074 int i;
2075 const uint32_t a= ((uint32_t*)(src-stride))[0];
2076 const uint32_t b= ((uint32_t*)(src-stride))[1];
2077 const uint32_t c= ((uint32_t*)(src-stride))[2];
2078 const uint32_t d= ((uint32_t*)(src-stride))[3];
115329f1 2079
0da71265
MN
2080 for(i=0; i<16; i++){
2081 ((uint32_t*)(src+i*stride))[0]= a;
2082 ((uint32_t*)(src+i*stride))[1]= b;
2083 ((uint32_t*)(src+i*stride))[2]= c;
2084 ((uint32_t*)(src+i*stride))[3]= d;
2085 }
2086}
2087
2088static void pred16x16_horizontal_c(uint8_t *src, int stride){
2089 int i;
2090
2091 for(i=0; i<16; i++){
2092 ((uint32_t*)(src+i*stride))[0]=
2093 ((uint32_t*)(src+i*stride))[1]=
2094 ((uint32_t*)(src+i*stride))[2]=
2095 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2096 }
2097}
2098
2099static void pred16x16_dc_c(uint8_t *src, int stride){
2100 int i, dc=0;
2101
2102 for(i=0;i<16; i++){
2103 dc+= src[-1+i*stride];
2104 }
115329f1 2105
0da71265
MN
2106 for(i=0;i<16; i++){
2107 dc+= src[i-stride];
2108 }
2109
2110 dc= 0x01010101*((dc + 16)>>5);
2111
2112 for(i=0; i<16; i++){
2113 ((uint32_t*)(src+i*stride))[0]=
2114 ((uint32_t*)(src+i*stride))[1]=
2115 ((uint32_t*)(src+i*stride))[2]=
2116 ((uint32_t*)(src+i*stride))[3]= dc;
2117 }
2118}
2119
2120static void pred16x16_left_dc_c(uint8_t *src, int stride){
2121 int i, dc=0;
2122
2123 for(i=0;i<16; i++){
2124 dc+= src[-1+i*stride];
2125 }
115329f1 2126
0da71265
MN
2127 dc= 0x01010101*((dc + 8)>>4);
2128
2129 for(i=0; i<16; i++){
2130 ((uint32_t*)(src+i*stride))[0]=
2131 ((uint32_t*)(src+i*stride))[1]=
2132 ((uint32_t*)(src+i*stride))[2]=
2133 ((uint32_t*)(src+i*stride))[3]= dc;
2134 }
2135}
2136
2137static void pred16x16_top_dc_c(uint8_t *src, int stride){
2138 int i, dc=0;
2139
2140 for(i=0;i<16; i++){
2141 dc+= src[i-stride];
2142 }
2143 dc= 0x01010101*((dc + 8)>>4);
2144
2145 for(i=0; i<16; i++){
2146 ((uint32_t*)(src+i*stride))[0]=
2147 ((uint32_t*)(src+i*stride))[1]=
2148 ((uint32_t*)(src+i*stride))[2]=
2149 ((uint32_t*)(src+i*stride))[3]= dc;
2150 }
2151}
2152
2153static void pred16x16_128_dc_c(uint8_t *src, int stride){
2154 int i;
2155
2156 for(i=0; i<16; i++){
2157 ((uint32_t*)(src+i*stride))[0]=
2158 ((uint32_t*)(src+i*stride))[1]=
2159 ((uint32_t*)(src+i*stride))[2]=
2160 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2161 }
2162}
2163
8b82a956 2164static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
30f73fc7
MN
2165 int i, j, k;
2166 int a;
2167 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2168 const uint8_t * const src0 = src+7-stride;
2169 const uint8_t *src1 = src+8*stride-1;
2170 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2171 int H = src0[1] - src0[-1];
2172 int V = src1[0] - src2[ 0];
2173 for(k=2; k<=8; ++k) {
2174 src1 += stride; src2 -= stride;
2175 H += k*(src0[k] - src0[-k]);
2176 V += k*(src1[0] - src2[ 0]);
2177 }
8b82a956
MN
2178 if(svq3){
2179 H = ( 5*(H/4) ) / 16;
2180 V = ( 5*(V/4) ) / 16;
2e26c8d2
MM
2181
2182 /* required for 100% accuracy */
2183 i = H; H = V; V = i;
8b82a956
MN
2184 }else{
2185 H = ( 5*H+32 ) >> 6;
2186 V = ( 5*V+32 ) >> 6;
2187 }
30f73fc7
MN
2188
2189 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2190 for(j=16; j>0; --j) {
2191 int b = a;
2192 a += V;
2193 for(i=-16; i<0; i+=4) {
2194 src[16+i] = cm[ (b ) >> 5 ];
2195 src[17+i] = cm[ (b+ H) >> 5 ];
2196 src[18+i] = cm[ (b+2*H) >> 5 ];
2197 src[19+i] = cm[ (b+3*H) >> 5 ];
2198 b += 4*H;
0da71265 2199 }
30f73fc7
MN
2200 src += stride;
2201 }
0da71265
MN
2202}
2203
8b82a956
MN
2204static void pred16x16_plane_c(uint8_t *src, int stride){
2205 pred16x16_plane_compat_c(src, stride, 0);
2206}
2207
0da71265
MN
2208static void pred8x8_vertical_c(uint8_t *src, int stride){
2209 int i;
2210 const uint32_t a= ((uint32_t*)(src-stride))[0];
2211 const uint32_t b= ((uint32_t*)(src-stride))[1];
115329f1 2212
0da71265
MN
2213 for(i=0; i<8; i++){
2214 ((uint32_t*)(src+i*stride))[0]= a;
2215 ((uint32_t*)(src+i*stride))[1]= b;
2216 }
2217}
2218
2219static void pred8x8_horizontal_c(uint8_t *src, int stride){
2220 int i;
2221
2222 for(i=0; i<8; i++){
2223 ((uint32_t*)(src+i*stride))[0]=
2224 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2225 }
2226}
2227
2228static void pred8x8_128_dc_c(uint8_t *src, int stride){
2229 int i;
2230
43efd19a 2231 for(i=0; i<8; i++){
115329f1 2232 ((uint32_t*)(src+i*stride))[0]=
0da71265
MN
2233 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2234 }
2235}
2236
2237static void pred8x8_left_dc_c(uint8_t *src, int stride){
2238 int i;
2239 int dc0, dc2;
2240
2241 dc0=dc2=0;
2242 for(i=0;i<4; i++){
2243 dc0+= src[-1+i*stride];
2244 dc2+= src[-1+(i+4)*stride];
2245 }
2246 dc0= 0x01010101*((dc0 + 2)>>2);
2247 dc2= 0x01010101*((dc2 + 2)>>2);
2248
2249 for(i=0; i<4; i++){
2250 ((uint32_t*)(src+i*stride))[0]=
2251 ((uint32_t*)(src+i*stride))[1]= dc0;
2252 }
2253 for(i=4; i<8; i++){
2254 ((uint32_t*)(src+i*stride))[0]=
2255 ((uint32_t*)(src+i*stride))[1]= dc2;
2256 }
2257}
2258
2259static void pred8x8_top_dc_c(uint8_t *src, int stride){
2260 int i;
2261 int dc0, dc1;
2262
2263 dc0=dc1=0;
2264 for(i=0;i<4; i++){
2265 dc0+= src[i-stride];
2266 dc1+= src[4+i-stride];
2267 }
2268 dc0= 0x01010101*((dc0 + 2)>>2);
2269 dc1= 0x01010101*((dc1 + 2)>>2);
2270
2271 for(i=0; i<4; i++){
2272 ((uint32_t*)(src+i*stride))[0]= dc0;
2273 ((uint32_t*)(src+i*stride))[1]= dc1;
2274 }
2275 for(i=4; i<8; i++){
2276 ((uint32_t*)(src+i*stride))[0]= dc0;
2277 ((uint32_t*)(src+i*stride))[1]= dc1;
2278 }
2279}
2280
2281
2282static void pred8x8_dc_c(uint8_t *src, int stride){
2283 int i;
2284 int dc0, dc1, dc2, dc3;
2285
2286 dc0=dc1=dc2=0;
2287 for(i=0;i<4; i++){
2288 dc0+= src[-1+i*stride] + src[i-stride];
2289 dc1+= src[4+i-stride];
2290 dc2+= src[-1+(i+4)*stride];
2291 }
2292 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2293 dc0= 0x01010101*((dc0 + 4)>>3);
2294 dc1= 0x01010101*((dc1 + 2)>>2);
2295 dc2= 0x01010101*((dc2 + 2)>>2);
2296
2297 for(i=0; i<4; i++){
2298 ((uint32_t*)(src+i*stride))[0]= dc0;
2299 ((uint32_t*)(src+i*stride))[1]= dc1;
2300 }
2301 for(i=4; i<8; i++){
2302 ((uint32_t*)(src+i*stride))[0]= dc2;
2303 ((uint32_t*)(src+i*stride))[1]= dc3;
2304 }
2305}
2306
2307static void pred8x8_plane_c(uint8_t *src, int stride){
30f73fc7
MN
2308 int j, k;
2309 int a;
2310 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2311 const uint8_t * const src0 = src+3-stride;
2312 const uint8_t *src1 = src+4*stride-1;
2313 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2314 int H = src0[1] - src0[-1];
2315 int V = src1[0] - src2[ 0];
2316 for(k=2; k<=4; ++k) {
2317 src1 += stride; src2 -= stride;
2318 H += k*(src0[k] - src0[-k]);
2319 V += k*(src1[0] - src2[ 0]);
2320 }
2321 H = ( 17*H+16 ) >> 5;
2322 V = ( 17*V+16 ) >> 5;
2323
2324 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2325 for(j=8; j>0; --j) {
2326 int b = a;
2327 a += V;
2328 src[0] = cm[ (b ) >> 5 ];
2329 src[1] = cm[ (b+ H) >> 5 ];
2330 src[2] = cm[ (b+2*H) >> 5 ];
2331 src[3] = cm[ (b+3*H) >> 5 ];
2332 src[4] = cm[ (b+4*H) >> 5 ];
2333 src[5] = cm[ (b+5*H) >> 5 ];
2334 src[6] = cm[ (b+6*H) >> 5 ];
2335 src[7] = cm[ (b+7*H) >> 5 ];
2336 src += stride;
2337 }
0da71265
MN
2338}
2339
43efd19a
LM
2340#define SRC(x,y) src[(x)+(y)*stride]
2341#define PL(y) \
2342 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2343#define PREDICT_8x8_LOAD_LEFT \
2344 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2345 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2346 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
bf4e3bd2 2347 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
43efd19a
LM
2348
2349#define PT(x) \
2350 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2351#define PREDICT_8x8_LOAD_TOP \
2352 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2353 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2354 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
bf4e3bd2 2355 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
fb3ca6ba 2356 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
43efd19a
LM
2357
2358#define PTR(x) \
2359 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2360#define PREDICT_8x8_LOAD_TOPRIGHT \
2361 int t8, t9, t10, t11, t12, t13, t14, t15; \
2362 if(has_topright) { \
2363 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2364 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2365 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2366
2367#define PREDICT_8x8_LOAD_TOPLEFT \
fb3ca6ba 2368 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
43efd19a
LM
2369
2370#define PREDICT_8x8_DC(v) \
2371 int y; \
2372 for( y = 0; y < 8; y++ ) { \
2373 ((uint32_t*)src)[0] = \
2374 ((uint32_t*)src)[1] = v; \
2375 src += stride; \
2376 }
2377
2378static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2379{
2380 PREDICT_8x8_DC(0x80808080);
2381}
2382static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2383{
2384 PREDICT_8x8_LOAD_LEFT;
2385 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2386 PREDICT_8x8_DC(dc);
2387}
2388static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2389{
2390 PREDICT_8x8_LOAD_TOP;
2391 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2392 PREDICT_8x8_DC(dc);
2393}
2394static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2395{
2396 PREDICT_8x8_LOAD_LEFT;
2397 PREDICT_8x8_LOAD_TOP;
2398 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2399 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2400 PREDICT_8x8_DC(dc);
2401}
2402static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2403{
2404 PREDICT_8x8_LOAD_LEFT;
2405#define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2406 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2407 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2408#undef ROW
2409}
2410static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2411{
2412 int y;
2413 PREDICT_8x8_LOAD_TOP;
2414 src[0] = t0;
2415 src[1] = t1;
2416 src[2] = t2;
2417 src[3] = t3;
2418 src[4] = t4;
2419 src[5] = t5;
2420 src[6] = t6;
2421 src[7] = t7;
2422 for( y = 1; y < 8; y++ )
2423 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2424}
2425static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2426{
2427 PREDICT_8x8_LOAD_TOP;
2428 PREDICT_8x8_LOAD_TOPRIGHT;
2429 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2430 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2431 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2432 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2433 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2434 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2435 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;
2436 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;
2437 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;
2438 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2439 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2440 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2441 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2442 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2443 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2444}
2445static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2446{
2447 PREDICT_8x8_LOAD_TOP;
2448 PREDICT_8x8_LOAD_LEFT;
2449 PREDICT_8x8_LOAD_TOPLEFT;
2450 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2451 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2452 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2453 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2454 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2455 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2456 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;
2457 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;
2458 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;
2459 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2460 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2461 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2462 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2463 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2464 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
115329f1 2465
43efd19a
LM
2466}
2467static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2468{
2469 PREDICT_8x8_LOAD_TOP;
2470 PREDICT_8x8_LOAD_LEFT;
2471 PREDICT_8x8_LOAD_TOPLEFT;
2472 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2473 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2474 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2475 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2476 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2477 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2478 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2479 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2480 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2481 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2482 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2483 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2484 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2485 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2486 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2487 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2488 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2489 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2490 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2491 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2492 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2493 SRC(7,0)= (t6 + t7 + 1) >> 1;
2494}
2495static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2496{
2497 PREDICT_8x8_LOAD_TOP;
2498 PREDICT_8x8_LOAD_LEFT;
2499 PREDICT_8x8_LOAD_TOPLEFT;
2500 SRC(0,7)= (l6 + l7 + 1) >> 1;
2501 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2502 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2503 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2504 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2505 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2506 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2507 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2508 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2509 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2510 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2511 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2512 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2513 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2514 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2515 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2516 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2517 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2518 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2519 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2520 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2521 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2522}
2523static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2524{
2525 PREDICT_8x8_LOAD_TOP;
2526 PREDICT_8x8_LOAD_TOPRIGHT;
2527 SRC(0,0)= (t0 + t1 + 1) >> 1;
2528 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2529 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2530 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2531 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2532 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2533 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2534 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2535 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2536 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2537 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2538 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2539 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2540 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2541 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2542 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2543 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2544 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2545 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2546 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2547 SRC(7,6)= (t10 + t11 + 1) >> 1;
2548 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2549}
2550static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2551{
2552 PREDICT_8x8_LOAD_LEFT;
2553 SRC(0,0)= (l0 + l1 + 1) >> 1;
2554 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2555 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2556 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2557 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2558 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2559 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2560 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2561 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2562 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2563 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2564 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2565 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2566 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2567 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2568 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2569 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2570 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2571}
2572#undef PREDICT_8x8_LOAD_LEFT
2573#undef PREDICT_8x8_LOAD_TOP
2574#undef PREDICT_8x8_LOAD_TOPLEFT
2575#undef PREDICT_8x8_LOAD_TOPRIGHT
2576#undef PREDICT_8x8_DC
2577#undef PTR
2578#undef PT
2579#undef PL
2580#undef SRC
2581
0da71265
MN
2582static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2583 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2584 int src_x_offset, int src_y_offset,
2585 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2586 MpegEncContext * const s = &h->s;
2587 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2588 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2589 const int luma_xy= (mx&3) + ((my&3)<<2);
2590 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2591 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2592 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2593 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2594 int extra_height= extra_width;
2595 int emu=0;
2596 const int full_mx= mx>>2;
2597 const int full_my= my>>2;
fbd312fd
LM
2598 const int pic_width = 16*s->mb_width;
2599 const int pic_height = 16*s->mb_height;
115329f1 2600
171c4076
LM
2601 if(!pic->data[0])
2602 return;
115329f1 2603
0da71265
MN
2604 if(mx&7) extra_width -= 3;
2605 if(my&7) extra_height -= 3;
115329f1
DB
2606
2607 if( full_mx < 0-extra_width
2608 || full_my < 0-extra_height
2609 || full_mx + 16/*FIXME*/ > pic_width + extra_width
fbd312fd
LM
2610 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2611 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
2612 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2613 emu=1;
2614 }
115329f1 2615
0da71265
MN
2616 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2617 if(!square){
2618 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2619 }
115329f1 2620
0da71265 2621 if(s->flags&CODEC_FLAG_GRAY) return;
115329f1 2622
0da71265 2623 if(emu){
fbd312fd 2624 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
2625 src_cb= s->edge_emu_buffer;
2626 }
2627 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2628
2629 if(emu){
fbd312fd 2630 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
2631 src_cr= s->edge_emu_buffer;
2632 }
2633 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2634}
2635
9f2d1b4f 2636static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
2637 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2638 int x_offset, int y_offset,
2639 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2640 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2641 int list0, int list1){
2642 MpegEncContext * const s = &h->s;
2643 qpel_mc_func *qpix_op= qpix_put;
2644 h264_chroma_mc_func chroma_op= chroma_put;
115329f1 2645
0da71265
MN
2646 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2647 dest_cb += x_offset + y_offset*s->uvlinesize;
2648 dest_cr += x_offset + y_offset*s->uvlinesize;
2649 x_offset += 8*s->mb_x;
2650 y_offset += 8*s->mb_y;
115329f1 2651
0da71265 2652 if(list0){
1924f3ce 2653 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
2654 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2655 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2656 qpix_op, chroma_op);
2657
2658 qpix_op= qpix_avg;
2659 chroma_op= chroma_avg;
2660 }
2661
2662 if(list1){
1924f3ce 2663 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
2664 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2665 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2666 qpix_op, chroma_op);
2667 }
2668}
2669
9f2d1b4f
LM
2670static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2671 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2672 int x_offset, int y_offset,
2673 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2674 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2675 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2676 int list0, int list1){
2677 MpegEncContext * const s = &h->s;
2678
2679 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2680 dest_cb += x_offset + y_offset*s->uvlinesize;
2681 dest_cr += x_offset + y_offset*s->uvlinesize;
2682 x_offset += 8*s->mb_x;
2683 y_offset += 8*s->mb_y;
115329f1 2684
9f2d1b4f
LM
2685 if(list0 && list1){
2686 /* don't optimize for luma-only case, since B-frames usually
2687 * use implicit weights => chroma too. */
2688 uint8_t *tmp_cb = s->obmc_scratchpad;
2689 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2690 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2691 int refn0 = h->ref_cache[0][ scan8[n] ];
2692 int refn1 = h->ref_cache[1][ scan8[n] ];
2693
2694 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2695 dest_y, dest_cb, dest_cr,
2696 x_offset, y_offset, qpix_put, chroma_put);
2697 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2698 tmp_y, tmp_cb, tmp_cr,
2699 x_offset, y_offset, qpix_put, chroma_put);
2700
2701 if(h->use_weight == 2){
2702 int weight0 = h->implicit_weight[refn0][refn1];
2703 int weight1 = 64 - weight0;
e8b56208
LM
2704 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0);
2705 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2706 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
9f2d1b4f
LM
2707 }else{
2708 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
115329f1 2709 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
e8b56208 2710 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
9f2d1b4f 2711 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2712 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
e8b56208 2713 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
9f2d1b4f 2714 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
115329f1 2715 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
e8b56208 2716 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
9f2d1b4f
LM
2717 }
2718 }else{
2719 int list = list1 ? 1 : 0;
2720 int refn = h->ref_cache[list][ scan8[n] ];
2721 Picture *ref= &h->ref_list[list][refn];
2722 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2723 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2724 qpix_put, chroma_put);
2725
2726 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2727 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2728 if(h->use_weight_chroma){
2729 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2730 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2731 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2732 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2733 }
2734 }
2735}
2736
2737static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2738 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2739 int x_offset, int y_offset,
2740 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2741 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
115329f1 2742 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
9f2d1b4f
LM
2743 int list0, int list1){
2744 if((h->use_weight==2 && list0 && list1
2745 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2746 || h->use_weight==1)
2747 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2748 x_offset, y_offset, qpix_put, chroma_put,
2749 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2750 else
2751 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2752 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2753}
2754
0da71265
MN
2755static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2756 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
2757 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2758 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 2759 MpegEncContext * const s = &h->s;
7bc9090a 2760 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 2761 const int mb_type= s->current_picture.mb_type[mb_xy];
115329f1 2762
0da71265 2763 assert(IS_INTER(mb_type));
115329f1 2764
0da71265
MN
2765 if(IS_16X16(mb_type)){
2766 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2767 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 2768 &weight_op[0], &weight_avg[0],
0da71265
MN
2769 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2770 }else if(IS_16X8(mb_type)){
2771 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2772 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2773 &weight_op[1], &weight_avg[1],
0da71265
MN
2774 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2775 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2776 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2777 &weight_op[1], &weight_avg[1],
0da71265
MN
2778 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2779 }else if(IS_8X16(mb_type)){
2780 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2781 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2782 &weight_op[2], &weight_avg[2],
0da71265
MN
2783 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2784 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2785 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2786 &weight_op[2], &weight_avg[2],
0da71265
MN
2787 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2788 }else{
2789 int i;
115329f1 2790
0da71265
MN
2791 assert(IS_8X8(mb_type));
2792
2793 for(i=0; i<4; i++){
2794 const int sub_mb_type= h->sub_mb_type[i];
2795 const int n= 4*i;
2796 int x_offset= (i&1)<<2;
2797 int y_offset= (i&2)<<1;
2798
2799 if(IS_SUB_8X8(sub_mb_type)){
2800 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2801 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2802 &weight_op[3], &weight_avg[3],
0da71265
MN
2803 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2804 }else if(IS_SUB_8X4(sub_mb_type)){
2805 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2806 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2807 &weight_op[4], &weight_avg[4],
0da71265
MN
2808 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2809 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2810 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2811 &weight_op[4], &weight_avg[4],
0da71265
MN
2812 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2813 }else if(IS_SUB_4X8(sub_mb_type)){
2814 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2815 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2816 &weight_op[5], &weight_avg[5],
0da71265
MN
2817 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2818 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2819 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2820 &weight_op[5], &weight_avg[5],
0da71265
MN
2821 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2822 }else{
2823 int j;
2824 assert(IS_SUB_4X4(sub_mb_type));
2825 for(j=0; j<4; j++){
2826 int sub_x_offset= x_offset + 2*(j&1);
2827 int sub_y_offset= y_offset + (j&2);
2828 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2829 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2830 &weight_op[6], &weight_avg[6],
0da71265
MN
2831 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2832 }
2833 }
2834 }
2835 }
2836}
2837
2838static void decode_init_vlc(H264Context *h){
2839 static int done = 0;
2840
2841 if (!done) {
2842 int i;
2843 done = 1;
2844
115329f1 2845 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
0da71265 2846 &chroma_dc_coeff_token_len [0], 1, 1,
073c2593 2847 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
0da71265
MN
2848
2849 for(i=0; i<4; i++){
115329f1 2850 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
0da71265 2851 &coeff_token_len [i][0], 1, 1,
073c2593 2852 &coeff_token_bits[i][0], 1, 1, 1);
0da71265
MN
2853 }
2854
2855 for(i=0; i<3; i++){
2856 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2857 &chroma_dc_total_zeros_len [i][0], 1, 1,
073c2593 2858 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2859 }
2860 for(i=0; i<15; i++){
115329f1 2861 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
0da71265 2862 &total_zeros_len [i][0], 1, 1,
073c2593 2863 &total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2864 }
2865
2866 for(i=0; i<6; i++){
115329f1 2867 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
0da71265 2868 &run_len [i][0], 1, 1,
073c2593 2869 &run_bits[i][0], 1, 1, 1);
0da71265 2870 }
115329f1 2871 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
0da71265 2872 &run_len [6][0], 1, 1,
073c2593 2873 &run_bits[6][0], 1, 1, 1);
0da71265
MN
2874 }
2875}
2876
2877/**
2878 * Sets the intra prediction function pointers.
2879 */
2880static void init_pred_ptrs(H264Context *h){
2881// MpegEncContext * const s = &h->s;
2882
2883 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2884 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2885 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2886 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2887 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2888 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2889 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2890 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2891 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2892 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2893 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2894 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2895
43efd19a
LM
2896 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
2897 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
2898 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
2899 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2900 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2901 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
2902 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
2903 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
2904 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
2905 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
2906 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
2907 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
2908
0da71265
MN
2909 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2910 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2911 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2912 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2913 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2914 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2915 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2916
2917 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2918 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2919 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2920 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2921 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2922 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2923 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2924}
2925
0da71265 2926static void free_tables(H264Context *h){
0da71265 2927 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
2928 av_freep(&h->chroma_pred_mode_table);
2929 av_freep(&h->cbp_table);
9e528114
LA
2930 av_freep(&h->mvd_table[0]);
2931 av_freep(&h->mvd_table[1]);
5ad984c9 2932 av_freep(&h->direct_table);
0da71265
MN
2933 av_freep(&h->non_zero_count);
2934 av_freep(&h->slice_table_base);
6ba71fc4
LLL
2935 av_freep(&h->top_borders[1]);
2936 av_freep(&h->top_borders[0]);
0da71265 2937 h->slice_table= NULL;
e5017ab8 2938
0da71265
MN
2939 av_freep(&h->mb2b_xy);
2940 av_freep(&h->mb2b8_xy);
9f2d1b4f
LM
2941
2942 av_freep(&h->s.obmc_scratchpad);
0da71265
MN
2943}
2944
239ea04c
LM
2945static void init_dequant8_coeff_table(H264Context *h){
2946 int i,q,x;
548a1c8a 2947 const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
239ea04c
LM
2948 h->dequant8_coeff[0] = h->dequant8_buffer[0];
2949 h->dequant8_coeff[1] = h->dequant8_buffer[1];
2950
2951 for(i=0; i<2; i++ ){
2952 if(i && !memcmp(h->pps.scaling_matrix8[0], h->pps.scaling_matrix8[1], 64*sizeof(uint8_t))){
2953 h->dequant8_coeff[1] = h->dequant8_buffer[0];
2954 break;
2955 }
2956
2957 for(q=0; q<52; q++){
2958 int shift = div6[q];
2959 int idx = rem6[q];
2960 for(x=0; x<64; x++)
548a1c8a
LM
2961 h->dequant8_coeff[i][q][transpose ? (x>>3)|((x&7)<<3) : x] =
2962 ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
2963 h->pps.scaling_matrix8[i][x]) << shift;
239ea04c
LM
2964 }
2965 }
2966}
2967
2968static void init_dequant4_coeff_table(H264Context *h){
2969 int i,j,q,x;
ab2e3e2c 2970 const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
239ea04c
LM
2971 for(i=0; i<6; i++ ){
2972 h->dequant4_coeff[i] = h->dequant4_buffer[i];
2973 for(j=0; j<i; j++){
2974 if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
2975 h->dequant4_coeff[i] = h->dequant4_buffer[j];
2976 break;
2977 }
2978 }
2979 if(j<i)
2980 continue;
2981
2982 for(q=0; q<52; q++){
2983 int shift = div6[q] + 2;
2984 int idx = rem6[q];
2985 for(x=0; x<16; x++)
ab2e3e2c
LM
2986 h->dequant4_coeff[i][q][transpose ? (x>>2)|((x<<2)&0xF) : x] =
2987 ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
239ea04c
LM
2988 h->pps.scaling_matrix4[i][x]) << shift;
2989 }
2990 }
2991}
2992
2993static void init_dequant_tables(H264Context *h){
2994 int i,x;
2995 init_dequant4_coeff_table(h);
2996 if(h->pps.transform_8x8_mode)
2997 init_dequant8_coeff_table(h);
2998 if(h->sps.transform_bypass){
2999 for(i=0; i<6; i++)
3000 for(x=0; x<16; x++)
3001 h->dequant4_coeff[i][0][x] = 1<<6;
3002 if(h->pps.transform_8x8_mode)
3003 for(i=0; i<2; i++)
3004 for(x=0; x<64; x++)
3005 h->dequant8_coeff[i][0][x] = 1<<6;
3006 }
3007}
3008
3009
0da71265
MN
3010/**
3011 * allocates tables.
3b66c4c5 3012 * needs width/height
0da71265
MN
3013 */
3014static int alloc_tables(H264Context *h){
3015 MpegEncContext * const s = &h->s;
7bc9090a 3016 const int big_mb_num= s->mb_stride * (s->mb_height+1);
239ea04c 3017 int x,y;
0da71265
MN
3018
3019 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 3020
53c05b1e 3021 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
0da71265 3022 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
6ba71fc4
LLL
3023 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
3024 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
5d0e4cb8 3025 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 3026
e5017ab8
LA
3027 if( h->pps.cabac ) {
3028 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
9e528114
LA
3029 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
3030 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
5ad984c9 3031 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8
LA
3032 }
3033
0da71265 3034 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
7bc9090a 3035 h->slice_table= h->slice_table_base + s->mb_stride + 1;
0da71265 3036
a55f20bd
LM
3037 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
3038 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
0da71265
MN
3039 for(y=0; y<s->mb_height; y++){
3040 for(x=0; x<s->mb_width; x++){
7bc9090a 3041 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
3042 const int b_xy = 4*x + 4*y*h->b_stride;
3043 const int b8_xy= 2*x + 2*y*h->b8_stride;
115329f1 3044
0da71265
MN
3045 h->mb2b_xy [mb_xy]= b_xy;
3046 h->mb2b8_xy[mb_xy]= b8_xy;
3047 }
3048 }
9f2d1b4f 3049
9c6221ae
GV
3050 s->obmc_scratchpad = NULL;
3051
56edbd81
LM
3052 if(!h->dequant4_coeff[0])
3053 init_dequant_tables(h);
3054
0da71265
MN
3055 return 0;
3056fail:
3057 free_tables(h);
3058 return -1;
3059}
3060
3061static void common_init(H264Context *h){
3062 MpegEncContext * const s = &h->s;
0da71265
MN
3063
3064 s->width = s->avctx->width;
3065 s->height = s->avctx->height;
3066 s->codec_id= s->avctx->codec->id;
115329f1 3067
0da71265
MN
3068 init_pred_ptrs(h);
3069
239ea04c 3070 h->dequant_coeff_pps= -1;
9a41c2c7 3071 s->unrestricted_mv=1;
0da71265 3072 s->decode=1; //FIXME
56edbd81
LM
3073
3074 memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
3075 memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
0da71265
MN
3076}
3077
3078static int decode_init(AVCodecContext *avctx){
3079 H264Context *h= avctx->priv_data;
3080 MpegEncContext * const s = &h->s;
3081
3edcacde 3082 MPV_decode_defaults(s);
115329f1 3083
0da71265
MN
3084 s->avctx = avctx;
3085 common_init(h);
3086
3087 s->out_format = FMT_H264;
3088 s->workaround_bugs= avctx->workaround_bugs;
3089
3090 // set defaults
0da71265
MN
3091// s->decode_mb= ff_h263_decode_mb;
3092 s->low_delay= 1;
3093 avctx->pix_fmt= PIX_FMT_YUV420P;
3094
3095 decode_init_vlc(h);
115329f1 3096
26165f99
MR
3097 if(avctx->extradata_size > 0 && avctx->extradata &&
3098 *(char *)avctx->extradata == 1){
4770b1b4
RT
3099 h->is_avc = 1;
3100 h->got_avcC = 0;
26165f99
MR
3101 } else {
3102 h->is_avc = 0;
4770b1b4
RT
3103 }
3104
0da71265
MN
3105 return 0;
3106}
3107
af8aa846 3108static int frame_start(H264Context *h){
0da71265
MN
3109 MpegEncContext * const s = &h->s;
3110 int i;
3111
af8aa846
MN
3112 if(MPV_frame_start(s, s->avctx) < 0)
3113 return -1;
0da71265 3114 ff_er_frame_start(s);
0da71265
MN
3115
3116 assert(s->linesize && s->uvlinesize);
3117
3118 for(i=0; i<16; i++){
3119 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 3120 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3121 }
3122 for(i=0; i<4; i++){
3123 h->block_offset[16+i]=
3124 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
3125 h->block_offset[24+16+i]=
3126 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
3127 }
3128
934b0821
LM
3129 /* can't be in alloc_tables because linesize isn't known there.
3130 * FIXME: redo bipred weight to not require extra buffer? */
3131 if(!s->obmc_scratchpad)
3132 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3133
0da71265 3134// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
af8aa846 3135 return 0;
0da71265
MN
3136}
3137
53c05b1e
MN
3138static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3139 MpegEncContext * const s = &h->s;
3140 int i;
115329f1 3141
53c05b1e
MN
3142 src_y -= linesize;
3143 src_cb -= uvlinesize;
3144 src_cr -= uvlinesize;
3145
3b66c4c5 3146 // There are two lines saved, the line above the the top macroblock of a pair,
6867a90b 3147 // and the line above the bottom macroblock
6ba71fc4 3148 h->left_border[0]= h->top_borders[0][s->mb_x][15];
53c05b1e
MN
3149 for(i=1; i<17; i++){
3150 h->left_border[i]= src_y[15+i* linesize];
3151 }
115329f1 3152
6ba71fc4
LLL
3153 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3154 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
53c05b1e
MN
3155
3156 if(!(s->flags&CODEC_FLAG_GRAY)){
6ba71fc4
LLL
3157 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3158 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
53c05b1e
MN
3159 for(i=1; i<9; i++){
3160 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3161 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3162 }
6ba71fc4
LLL
3163 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3164 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
53c05b1e
MN
3165 }
3166}
3167
3168static 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){
3169 MpegEncContext * const s = &h->s;
3170 int temp8, i;
3171 uint64_t temp64;
d89dc06a
LM
3172 int deblock_left = (s->mb_x > 0);
3173 int deblock_top = (s->mb_y > 0);
53c05b1e
MN
3174
3175 src_y -= linesize + 1;
3176 src_cb -= uvlinesize + 1;
3177 src_cr -= uvlinesize + 1;
3178
3179#define XCHG(a,b,t,xchg)\
3180t= a;\
3181if(xchg)\
3182 a= b;\
3183b= t;
d89dc06a
LM
3184
3185 if(deblock_left){
3186 for(i = !deblock_top; i<17; i++){
3187 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3188 }
3189 }
3190
3191 if(deblock_top){
6ba71fc4
LLL
3192 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3193 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
cad4368a 3194 if(s->mb_x+1 < s->mb_width){
43efd19a
LM
3195 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3196 }
53c05b1e 3197 }
53c05b1e
MN
3198
3199 if(!(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
3200 if(deblock_left){
3201 for(i = !deblock_top; i<9; i++){
3202 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3203 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3204 }
3205 }
3206 if(deblock_top){
6ba71fc4
LLL
3207 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3208 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3209 }
3210 }
3211}
3212
3213static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3214 MpegEncContext * const s = &h->s;
3215 int i;
115329f1 3216
6ba71fc4
LLL
3217 src_y -= 2 * linesize;
3218 src_cb -= 2 * uvlinesize;
3219 src_cr -= 2 * uvlinesize;
3220
3b66c4c5 3221 // There are two lines saved, the line above the the top macroblock of a pair,
6ba71fc4
LLL
3222 // and the line above the bottom macroblock
3223 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3224 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3225 for(i=2; i<34; i++){
3226 h->left_border[i]= src_y[15+i* linesize];
3227 }
115329f1 3228
6ba71fc4
LLL
3229 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3230 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3231 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3232 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3233
3234 if(!(s->flags&CODEC_FLAG_GRAY)){
3235 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3236 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3237 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3238 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3239 for(i=2; i<18; i++){
3240 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3241 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3242 }
3243 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3244 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3245 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3246 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3247 }
3248}
3249
3250static 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){
3251 MpegEncContext * const s = &h->s;
3252 int temp8, i;
3253 uint64_t temp64;
3254 int deblock_left = (s->mb_x > 0);
3255 int deblock_top = (s->mb_y > 0);
3256
3257 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);
3258
3259 src_y -= 2 * linesize + 1;
3260 src_cb -= 2 * uvlinesize + 1;
3261 src_cr -= 2 * uvlinesize + 1;
3262
3263#define XCHG(a,b,t,xchg)\
3264t= a;\
3265if(xchg)\
3266 a= b;\
3267b= t;
3268
3269 if(deblock_left){
3270 for(i = (!deblock_top)<<1; i<34; i++){
3271 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3272 }
3273 }
3274
3275 if(deblock_top){
3276 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3277 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3278 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3279 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3280 }
3281
3282 if(!(s->flags&CODEC_FLAG_GRAY)){
3283 if(deblock_left){
3284 for(i = (!deblock_top) << 1; i<18; i++){
3285 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
3286 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3287 }
3288 }
3289 if(deblock_top){
3290 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3291 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3292 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3293 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
53c05b1e 3294 }
53c05b1e
MN
3295 }
3296}
3297
0da71265
MN
3298static void hl_decode_mb(H264Context *h){
3299 MpegEncContext * const s = &h->s;
3300 const int mb_x= s->mb_x;
3301 const int mb_y= s->mb_y;
7bc9090a 3302 const int mb_xy= mb_x + mb_y*s->mb_stride;
0da71265
MN
3303 const int mb_type= s->current_picture.mb_type[mb_xy];
3304 uint8_t *dest_y, *dest_cb, *dest_cr;
3305 int linesize, uvlinesize /*dct_offset*/;
3306 int i;
6867a90b 3307 int *block_offset = &h->block_offset[0];
6ba71fc4 3308 const unsigned int bottom = mb_y & 1;
36940eca
LM
3309 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3310 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15 3311 void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
0da71265
MN
3312
3313 if(!s->decode)
3314 return;
3315
0da71265
MN
3316 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3317 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3318 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3319
3320 if (h->mb_field_decoding_flag) {
3321 linesize = s->linesize * 2;
3322 uvlinesize = s->uvlinesize * 2;
6867a90b 3323 block_offset = &h->block_offset[24];
0da71265
MN
3324 if(mb_y&1){ //FIXME move out of this func?
3325 dest_y -= s->linesize*15;
6867a90b
LLL
3326 dest_cb-= s->uvlinesize*7;
3327 dest_cr-= s->uvlinesize*7;
0da71265
MN
3328 }
3329 } else {
3330 linesize = s->linesize;
3331 uvlinesize = s->uvlinesize;
3332// dct_offset = s->linesize * 16;
3333 }
115329f1 3334
ef9d1d15
LM
3335 if(transform_bypass){
3336 idct_dc_add =
3337 idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
3338 }else if(IS_8x8DCT(mb_type)){
3339 idct_dc_add = s->dsp.h264_idct8_dc_add;
3340 idct_add = s->dsp.h264_idct8_add;
3341 }else{
3342 idct_dc_add = s->dsp.h264_idct_dc_add;
3343 idct_add = s->dsp.h264_idct_add;
3344 }
0da71265 3345
6fbcaaa0
LLL
3346 if (IS_INTRA_PCM(mb_type)) {
3347 unsigned int x, y;
3348
3349 // The pixels are stored in h->mb array in the same order as levels,
3350 // copy them in output in the correct order.
3351 for(i=0; i<16; i++) {
3352 for (y=0; y<4; y++) {
3353 for (x=0; x<4; x++) {
6867a90b 3354 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3355 }
3356 }
3357 }
3358 for(i=16; i<16+4; i++) {
3359 for (y=0; y<4; y++) {
3360 for (x=0; x<4; x++) {
6867a90b 3361 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3362 }
3363 }
3364 }
3365 for(i=20; i<20+4; i++) {
3366 for (y=0; y<4; y++) {
3367 for (x=0; x<4; x++) {
6867a90b 3368 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
3369 }
3370 }
3371 }
e7e09b49
LLL
3372 } else {
3373 if(IS_INTRA(mb_type)){
6ba71fc4
LLL
3374 if(h->deblocking_filter) {
3375 if (h->mb_aff_frame) {
3376 if (!bottom)
3377 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3378 } else {
3379 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3380 }
3381 }
53c05b1e 3382
e7e09b49
LLL
3383 if(!(s->flags&CODEC_FLAG_GRAY)){
3384 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3385 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3386 }
0da71265 3387
e7e09b49
LLL
3388 if(IS_INTRA4x4(mb_type)){
3389 if(!s->encoding){
43efd19a
LM
3390 if(IS_8x8DCT(mb_type)){
3391 for(i=0; i<16; i+=4){
3392 uint8_t * const ptr= dest_y + block_offset[i];
3393 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3394 const int nnz = h->non_zero_count_cache[ scan8[i] ];
43efd19a
LM
3395 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3396 (h->topright_samples_available<<(i+1))&0x8000, linesize);
ef9d1d15
LM
3397 if(nnz){
3398 if(nnz == 1 && h->mb[i*16])
3399 idct_dc_add(ptr, h->mb + i*16, linesize);
3400 else
3401 idct_add(ptr, h->mb + i*16, linesize);
3402 }
43efd19a
LM
3403 }
3404 }else
e7e09b49 3405 for(i=0; i<16; i++){
6867a90b 3406 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
3407 uint8_t *topright;
3408 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
ef9d1d15 3409 int nnz, tr;
e7e09b49
LLL
3410
3411 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3412 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6867a90b 3413 assert(mb_y || linesize <= block_offset[i]);
e7e09b49
LLL
3414 if(!topright_avail){
3415 tr= ptr[3 - linesize]*0x01010101;
3416 topright= (uint8_t*) &tr;
115329f1 3417 }else
e7e09b49 3418 topright= ptr + 4 - linesize;
a9799653 3419 }else
e7e09b49
LLL
3420 topright= NULL;
3421
3422 h->pred4x4[ dir ](ptr, topright, linesize);
ef9d1d15
LM
3423 nnz = h->non_zero_count_cache[ scan8[i] ];
3424 if(nnz){
3425 if(s->codec_id == CODEC_ID_H264){
3426 if(nnz == 1 && h->mb[i*16])
3427 idct_dc_add(ptr, h->mb + i*16, linesize);
3428 else
3429 idct_add(ptr, h->mb + i*16, linesize);
3430 }else
e7e09b49
LLL
3431 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3432 }
8b82a956 3433 }
0da71265 3434 }
e7e09b49
LLL
3435 }else{
3436 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
36940eca
LM
3437 if(s->codec_id == CODEC_ID_H264){
3438 if(!transform_bypass)
239ea04c 3439 h264_luma_dc_dequant_idct_c(h->mb, s->qscale, h->dequant4_coeff[IS_INTRA(mb_type) ? 0:3][s->qscale][0]);
36940eca 3440 }else
e7e09b49 3441 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
0da71265 3442 }
6ba71fc4
LLL
3443 if(h->deblocking_filter) {
3444 if (h->mb_aff_frame) {
3445 if (bottom) {
3446 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3447 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3448 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3449 s->mb_y--;
3450 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3451 s->mb_y++;
3452 }
3453 } else {
3454 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3455 }
3456 }
e7e09b49
LLL
3457 }else if(s->codec_id == CODEC_ID_H264){
3458 hl_motion(h, dest_y, dest_cb, dest_cr,
115329f1 3459 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
e7e09b49
LLL
3460 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3461 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
0da71265 3462 }
e7e09b49
LLL
3463
3464
3465 if(!IS_INTRA4x4(mb_type)){
3466 if(s->codec_id == CODEC_ID_H264){
ef9d1d15
LM
3467 if(IS_INTRA16x16(mb_type)){
3468 for(i=0; i<16; i++){
3469 if(h->non_zero_count_cache[ scan8[i] ])
3470 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3471 else if(h->mb[i*16])
3472 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3473 }
3474 }else{
3475 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3476 for(i=0; i<16; i+=di){
3477 int nnz = h->non_zero_count_cache[ scan8[i] ];
3478 if(nnz){
3479 if(nnz==1 && h->mb[i*16])
3480 idct_dc_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3481 else
3482 idct_add(dest_y + block_offset[i], h->mb + i*16, linesize);
3483 }
e7e09b49 3484 }
4704097a 3485 }
e7e09b49
LLL
3486 }else{
3487 for(i=0; i<16; i++){
3488 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
6867a90b 3489 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
3490 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3491 }
4704097a 3492 }
0da71265
MN
3493 }
3494 }
0da71265 3495
e7e09b49 3496 if(!(s->flags&CODEC_FLAG_GRAY)){
ef9d1d15
LM
3497 uint8_t *dest[2] = {dest_cb, dest_cr};