Mercurial > libavcodec.hg
annotate rv34.c @ 10118:4dafbd9c8918 libavcodec
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
architectures since stride is multiple of 4 and not of 8, so split
fill_rectangle() calls to operate on 32-bit words instead of 64-bit ones.
author | kostya |
---|---|
date | Sun, 30 Aug 2009 06:30:53 +0000 |
parents | d0f2f4e4436c |
children | e37de8b92b10 |
rev | line source |
---|---|
6026 | 1 /* |
2 * RV30/40 decoder common data | |
3 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov | |
4 * | |
5 * This file is part of FFmpeg. | |
6 * | |
7 * FFmpeg is free software; you can redistribute it and/or | |
8 * modify it under the terms of the GNU Lesser General Public | |
9 * License as published by the Free Software Foundation; either | |
10 * version 2.1 of the License, or (at your option) any later version. | |
11 * | |
12 * FFmpeg is distributed in the hope that it will be useful, | |
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
18 * License along with FFmpeg; if not, write to the Free Software | |
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 */ | |
21 | |
22 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8693
diff
changeset
|
23 * @file libavcodec/rv34.c |
6026 | 24 * RV30/40 decoder common data |
25 */ | |
26 | |
27 #include "avcodec.h" | |
28 #include "dsputil.h" | |
29 #include "mpegvideo.h" | |
30 #include "golomb.h" | |
8627
d6bab465b82c
moves mid_pred() into mathops.h (with arch specific code split by directory)
aurel
parents:
8504
diff
changeset
|
31 #include "mathops.h" |
6026 | 32 #include "rectangle.h" |
33 | |
34 #include "rv34vlc.h" | |
35 #include "rv34data.h" | |
36 #include "rv34.h" | |
37 | |
38 //#define DEBUG | |
39 | |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
40 #define ZERO8x2(dst, stride) \ |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
41 fill_rectangle(dst, 1, 2, stride, 0, 4); \ |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
42 fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4); \ |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
43 |
6026 | 44 /** translation of RV30/40 macroblock types to lavc ones */ |
45 static const int rv34_mb_type_to_lavc[12] = { | |
46 MB_TYPE_INTRA, | |
8036
9b98b2261938
Mark some of RV3/4 block types as having separate DC subblock
kostya
parents:
6750
diff
changeset
|
47 MB_TYPE_INTRA16x16 | MB_TYPE_SEPARATE_DC, |
6026 | 48 MB_TYPE_16x16 | MB_TYPE_L0, |
49 MB_TYPE_8x8 | MB_TYPE_L0, | |
50 MB_TYPE_16x16 | MB_TYPE_L0, | |
51 MB_TYPE_16x16 | MB_TYPE_L1, | |
52 MB_TYPE_SKIP, | |
53 MB_TYPE_DIRECT2 | MB_TYPE_16x16, | |
54 MB_TYPE_16x8 | MB_TYPE_L0, | |
55 MB_TYPE_8x16 | MB_TYPE_L0, | |
56 MB_TYPE_16x16 | MB_TYPE_L0L1, | |
8036
9b98b2261938
Mark some of RV3/4 block types as having separate DC subblock
kostya
parents:
6750
diff
changeset
|
57 MB_TYPE_16x16 | MB_TYPE_L0 | MB_TYPE_SEPARATE_DC |
6026 | 58 }; |
59 | |
60 | |
61 static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES]; | |
62 | |
63 /** | |
64 * @defgroup vlc RV30/40 VLC generating functions | |
65 * @{ | |
66 */ | |
67 | |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
68 static const int table_offs[] = { |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
69 0, 1818, 3622, 4144, 4698, 5234, 5804, 5868, 5900, 5932, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
70 5996, 6252, 6316, 6348, 6380, 7674, 8944, 10274, 11668, 12250, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
71 14060, 15846, 16372, 16962, 17512, 18148, 18180, 18212, 18244, 18308, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
72 18564, 18628, 18660, 18692, 20036, 21314, 22648, 23968, 24614, 26384, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
73 28190, 28736, 29366, 29938, 30608, 30640, 30672, 30704, 30768, 31024, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
74 31088, 31120, 31184, 32570, 33898, 35236, 36644, 37286, 39020, 40802, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
75 41368, 42052, 42692, 43348, 43380, 43412, 43444, 43476, 43604, 43668, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
76 43700, 43732, 45100, 46430, 47778, 49160, 49802, 51550, 53340, 53972, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
77 54648, 55348, 55994, 56122, 56154, 56186, 56218, 56346, 56410, 56442, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
78 56474, 57878, 59290, 60636, 62036, 62682, 64460, 64524, 64588, 64716, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
79 64844, 66076, 67466, 67978, 68542, 69064, 69648, 70296, 72010, 72074, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
80 72138, 72202, 72330, 73572, 74936, 75454, 76030, 76566, 77176, 77822, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
81 79582, 79646, 79678, 79742, 79870, 81180, 82536, 83064, 83672, 84242, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
82 84934, 85576, 87384, 87448, 87480, 87544, 87672, 88982, 90340, 90902, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
83 91598, 92182, 92846, 93488, 95246, 95278, 95310, 95374, 95502, 96878, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
84 98266, 98848, 99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
85 103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
86 111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592 |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
87 }; |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
88 |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
89 static VLC_TYPE table_data[117592][2]; |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
90 |
6026 | 91 /** |
92 * Generate VLC from codeword lengths. | |
93 * @param bits codeword lengths (zeroes are accepted) | |
94 * @param size length of input data | |
8684 | 95 * @param vlc output VLC |
6026 | 96 * @param insyms symbols for input codes (NULL for default ones) |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
97 * @param num VLC table number (for static initialization) |
6026 | 98 */ |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
99 static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms, |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
100 const int num) |
6026 | 101 { |
102 int i; | |
103 int counts[17] = {0}, codes[17]; | |
104 uint16_t cw[size], syms[size]; | |
105 uint8_t bits2[size]; | |
106 int maxbits = 0, realsize = 0; | |
107 | |
108 for(i = 0; i < size; i++){ | |
109 if(bits[i]){ | |
110 bits2[realsize] = bits[i]; | |
111 syms[realsize] = insyms ? insyms[i] : i; | |
112 realsize++; | |
113 maxbits = FFMAX(maxbits, bits[i]); | |
114 counts[bits[i]]++; | |
115 } | |
116 } | |
117 | |
118 codes[0] = 0; | |
119 for(i = 0; i < 16; i++) | |
120 codes[i+1] = (codes[i] + counts[i]) << 1; | |
121 for(i = 0; i < realsize; i++) | |
122 cw[i] = codes[bits2[i]]++; | |
123 | |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
124 vlc->table = &table_data[table_offs[num]]; |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
125 vlc->table_allocated = table_offs[num + 1] - table_offs[num]; |
6026 | 126 init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize, |
127 bits2, 1, 1, | |
128 cw, 2, 2, | |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
129 syms, 2, 2, INIT_VLC_USE_NEW_STATIC); |
6026 | 130 } |
131 | |
132 /** | |
133 * Initialize all tables. | |
134 */ | |
8693
18737839ed27
Add missing void keyword to parameterless function declarations.
diego
parents:
8684
diff
changeset
|
135 static av_cold void rv34_init_tables(void) |
6026 | 136 { |
137 int i, j, k; | |
138 | |
139 for(i = 0; i < NUM_INTRA_TABLES; i++){ | |
140 for(j = 0; j < 2; j++){ | |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
141 rv34_gen_vlc(rv34_table_intra_cbppat [i][j], CBPPAT_VLC_SIZE, &intra_vlcs[i].cbppattern[j], NULL, 19*i + 0 + j); |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
142 rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j); |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
143 rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j], NULL, 19*i + 4 + j); |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
144 for(k = 0; k < 4; k++){ |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
145 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2], CBP_VLC_SIZE, &intra_vlcs[i].cbp[j][k], rv34_cbp_code, 19*i + 6 + j*4 + k); |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
146 } |
6026 | 147 } |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
148 for(j = 0; j < 4; j++){ |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
149 rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j); |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
150 } |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
151 rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18); |
6026 | 152 } |
153 | |
154 for(i = 0; i < NUM_INTER_TABLES; i++){ | |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
155 rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95); |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
156 for(j = 0; j < 4; j++){ |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
157 rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j); |
9520
c9c23894f4ae
Make RV3/4 VLC tables use new static initialization method
kostya
parents:
9355
diff
changeset
|
158 } |
6026 | 159 for(j = 0; j < 2; j++){ |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
160 rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j], NULL, i*12 + 100 + j); |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
161 rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j); |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
162 rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j], NULL, i*12 + 104 + j); |
6026 | 163 } |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
164 rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106); |
6026 | 165 } |
166 } | |
167 | |
168 /** @} */ // vlc group | |
169 | |
170 | |
171 /** | |
172 * @defgroup transform RV30/40 inverse transform functions | |
173 * @{ | |
174 */ | |
175 | |
176 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block) | |
177 { | |
178 int i; | |
179 | |
180 for(i=0; i<4; i++){ | |
181 const int z0= 13*(block[i+8*0] + block[i+8*2]); | |
182 const int z1= 13*(block[i+8*0] - block[i+8*2]); | |
183 const int z2= 7* block[i+8*1] - 17*block[i+8*3]; | |
184 const int z3= 17* block[i+8*1] + 7*block[i+8*3]; | |
185 | |
186 temp[4*i+0]= z0+z3; | |
187 temp[4*i+1]= z1+z2; | |
188 temp[4*i+2]= z1-z2; | |
189 temp[4*i+3]= z0-z3; | |
190 } | |
191 } | |
192 | |
193 /** | |
194 * Real Video 3.0/4.0 inverse transform | |
195 * Code is almost the same as in SVQ3, only scaling is different. | |
196 */ | |
197 static void rv34_inv_transform(DCTELEM *block){ | |
198 int temp[16]; | |
199 int i; | |
200 | |
201 rv34_row_transform(temp, block); | |
202 | |
203 for(i=0; i<4; i++){ | |
204 const int z0= 13*(temp[4*0+i] + temp[4*2+i]) + 0x200; | |
205 const int z1= 13*(temp[4*0+i] - temp[4*2+i]) + 0x200; | |
206 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i]; | |
207 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i]; | |
208 | |
209 block[i*8+0]= (z0 + z3)>>10; | |
210 block[i*8+1]= (z1 + z2)>>10; | |
211 block[i*8+2]= (z1 - z2)>>10; | |
212 block[i*8+3]= (z0 - z3)>>10; | |
213 } | |
214 | |
215 } | |
216 | |
217 /** | |
218 * RealVideo 3.0/4.0 inverse transform for DC block | |
219 * | |
220 * Code is almost the same as rv34_inv_transform() | |
221 * but final coefficients are multiplied by 1.5 and have no rounding. | |
222 */ | |
223 static void rv34_inv_transform_noround(DCTELEM *block){ | |
224 int temp[16]; | |
225 int i; | |
226 | |
227 rv34_row_transform(temp, block); | |
228 | |
229 for(i=0; i<4; i++){ | |
230 const int z0= 13*(temp[4*0+i] + temp[4*2+i]); | |
231 const int z1= 13*(temp[4*0+i] - temp[4*2+i]); | |
232 const int z2= 7* temp[4*1+i] - 17*temp[4*3+i]; | |
233 const int z3= 17* temp[4*1+i] + 7*temp[4*3+i]; | |
234 | |
235 block[i*8+0]= ((z0 + z3)*3)>>11; | |
236 block[i*8+1]= ((z1 + z2)*3)>>11; | |
237 block[i*8+2]= ((z1 - z2)*3)>>11; | |
238 block[i*8+3]= ((z0 - z3)*3)>>11; | |
239 } | |
240 | |
241 } | |
242 | |
243 /** @} */ // transform | |
244 | |
245 | |
246 /** | |
247 * @defgroup block RV30/40 4x4 block decoding functions | |
248 * @{ | |
249 */ | |
250 | |
251 /** | |
252 * Decode coded block pattern. | |
253 */ | |
254 static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table) | |
255 { | |
256 int pattern, code, cbp=0; | |
257 int ones; | |
258 static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000}; | |
259 static const int shifts[4] = { 0, 2, 8, 10 }; | |
8721
42e258f2b931
Silence two pointer assignment compiler warnings in rv34.c
kostya
parents:
8718
diff
changeset
|
260 const int *curshift = shifts; |
6026 | 261 int i, t, mask; |
262 | |
263 code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2); | |
264 pattern = code & 0xF; | |
265 code >>= 4; | |
266 | |
267 ones = rv34_count_ones[pattern]; | |
268 | |
269 for(mask = 8; mask; mask >>= 1, curshift++){ | |
270 if(pattern & mask) | |
271 cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0]; | |
272 } | |
273 | |
274 for(i = 0; i < 4; i++){ | |
275 t = modulo_three_table[code][i]; | |
276 if(t == 1) | |
277 cbp |= cbp_masks[get_bits1(gb)] << i; | |
278 if(t == 2) | |
279 cbp |= cbp_masks[2] << i; | |
280 } | |
281 return cbp; | |
282 } | |
283 | |
284 /** | |
285 * Get one coefficient value from the bistream and store it. | |
286 */ | |
287 static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc) | |
288 { | |
289 if(coef){ | |
290 if(coef == esc){ | |
291 coef = get_vlc2(gb, vlc->table, 9, 2); | |
292 if(coef > 23){ | |
293 coef -= 23; | |
294 coef = 22 + ((1 << coef) | get_bits(gb, coef)); | |
295 } | |
296 coef += esc; | |
297 } | |
298 if(get_bits1(gb)) | |
299 coef = -coef; | |
300 *dst = coef; | |
301 } | |
302 } | |
303 | |
304 /** | |
305 * Decode 2x2 subblock of coefficients. | |
306 */ | |
307 static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc) | |
308 { | |
309 int coeffs[4]; | |
310 | |
311 coeffs[0] = modulo_three_table[code][0]; | |
312 coeffs[1] = modulo_three_table[code][1]; | |
313 coeffs[2] = modulo_three_table[code][2]; | |
314 coeffs[3] = modulo_three_table[code][3]; | |
315 decode_coeff(dst , coeffs[0], 3, gb, vlc); | |
316 if(is_block2){ | |
317 decode_coeff(dst+8, coeffs[1], 2, gb, vlc); | |
318 decode_coeff(dst+1, coeffs[2], 2, gb, vlc); | |
319 }else{ | |
320 decode_coeff(dst+1, coeffs[1], 2, gb, vlc); | |
321 decode_coeff(dst+8, coeffs[2], 2, gb, vlc); | |
322 } | |
323 decode_coeff(dst+9, coeffs[3], 2, gb, vlc); | |
324 } | |
325 | |
326 /** | |
327 * Decode coefficients for 4x4 block. | |
328 * | |
329 * This is done by filling 2x2 subblocks with decoded coefficients | |
330 * in this order (the same for subblocks and subblock coefficients): | |
331 * o--o | |
332 * / | |
333 * / | |
334 * o--o | |
335 */ | |
336 | |
337 static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc) | |
338 { | |
339 int code, pattern; | |
340 | |
341 code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2); | |
342 | |
343 pattern = code & 0x7; | |
344 | |
345 code >>= 3; | |
346 decode_subblock(dst, code, 0, gb, &rvlc->coefficient); | |
347 | |
348 if(pattern & 4){ | |
349 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2); | |
350 decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient); | |
351 } | |
352 if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block | |
353 code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2); | |
354 decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient); | |
355 } | |
356 if(pattern & 1){ | |
357 code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2); | |
358 decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient); | |
359 } | |
360 | |
361 } | |
362 | |
363 /** | |
364 * Dequantize ordinary 4x4 block. | |
365 * @todo optimize | |
366 */ | |
367 static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q) | |
368 { | |
369 int i, j; | |
370 | |
371 block[0] = (block[0] * Qdc + 8) >> 4; | |
372 for(i = 0; i < 4; i++) | |
373 for(j = !i; j < 4; j++) | |
374 block[j + i*8] = (block[j + i*8] * Q + 8) >> 4; | |
375 } | |
376 | |
377 /** | |
378 * Dequantize 4x4 block of DC values for 16x16 macroblock. | |
379 * @todo optimize | |
380 */ | |
381 static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q) | |
382 { | |
383 int i; | |
384 | |
385 for(i = 0; i < 3; i++) | |
386 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4; | |
387 for(; i < 16; i++) | |
388 block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4; | |
389 } | |
390 /** @} */ //block functions | |
391 | |
392 | |
393 /** | |
394 * @defgroup bitstream RV30/40 bitstream parsing | |
395 * @{ | |
396 */ | |
397 | |
398 /** | |
399 * Decode starting slice position. | |
400 * @todo Maybe replace with ff_h263_decode_mba() ? | |
401 */ | |
402 int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size) | |
403 { | |
404 int i; | |
405 for(i = 0; i < 5; i++) | |
8445
8621deaff8dc
352l: correct calculating number of bits for storing macroblock offset in RV3/4
kostya
parents:
8371
diff
changeset
|
406 if(rv34_mb_max_sizes[i] >= mb_size - 1) |
6026 | 407 break; |
408 return rv34_mb_bits_sizes[i]; | |
409 } | |
410 | |
411 /** | |
412 * Select VLC set for decoding from current quantizer, modifier and frame type. | |
413 */ | |
414 static inline RV34VLC* choose_vlc_set(int quant, int mod, int type) | |
415 { | |
416 if(mod == 2 && quant < 19) quant += 10; | |
417 else if(mod && quant < 26) quant += 5; | |
418 return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]] | |
419 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]]; | |
420 } | |
421 | |
422 /** | |
423 * Decode quantizer difference and return modified quantizer. | |
424 */ | |
425 static inline int rv34_decode_dquant(GetBitContext *gb, int quant) | |
426 { | |
427 if(get_bits1(gb)) | |
428 return rv34_dquant_tab[get_bits1(gb)][quant]; | |
429 else | |
430 return get_bits(gb, 5); | |
431 } | |
432 | |
433 /** @} */ //bitstream functions | |
434 | |
435 /** | |
436 * @defgroup mv motion vector related code (prediction, reconstruction, motion compensation) | |
437 * @{ | |
438 */ | |
439 | |
440 /** macroblock partition width in 8x8 blocks */ | |
441 static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 }; | |
442 | |
443 /** macroblock partition height in 8x8 blocks */ | |
444 static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 }; | |
445 | |
446 /** availability index for subblocks */ | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
447 static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 }; |
6026 | 448 |
449 /** | |
450 * motion vector prediction | |
451 * | |
452 * Motion prediction performed for the block by using median prediction of | |
453 * motion vectors from the left, top and right top blocks but in corner cases | |
454 * some other vectors may be used instead. | |
455 */ | |
456 static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no) | |
457 { | |
458 MpegEncContext *s = &r->s; | |
459 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; | |
460 int A[2] = {0}, B[2], C[2]; | |
461 int i, j; | |
462 int mx, my; | |
463 int avail_index = avail_indexes[subblock_no]; | |
464 int c_off = part_sizes_w[block_type]; | |
465 | |
466 mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride; | |
467 if(subblock_no == 3) | |
468 c_off = -1; | |
469 | |
470 if(r->avail_cache[avail_index - 1]){ | |
471 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0]; | |
472 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1]; | |
473 } | |
474 if(r->avail_cache[avail_index - 4]){ | |
475 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0]; | |
476 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1]; | |
477 }else{ | |
478 B[0] = A[0]; | |
479 B[1] = A[1]; | |
480 } | |
481 if(!r->avail_cache[avail_index - 4 + c_off]){ | |
482 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){ | |
483 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0]; | |
484 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1]; | |
485 }else{ | |
486 C[0] = A[0]; | |
487 C[1] = A[1]; | |
488 } | |
489 }else{ | |
490 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0]; | |
491 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1]; | |
492 } | |
493 mx = mid_pred(A[0], B[0], C[0]); | |
494 my = mid_pred(A[1], B[1], C[1]); | |
495 mx += r->dmv[dmv_no][0]; | |
496 my += r->dmv[dmv_no][1]; | |
497 for(j = 0; j < part_sizes_h[block_type]; j++){ | |
498 for(i = 0; i < part_sizes_w[block_type]; i++){ | |
499 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx; | |
500 s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my; | |
501 } | |
502 } | |
503 } | |
504 | |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
505 #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF) |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
506 |
6026 | 507 /** |
6096
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
508 * Calculate motion vector component that should be added for direct blocks. |
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
509 */ |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
510 static int calc_add_mv(RV34DecContext *r, int dir, int val) |
6096
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
511 { |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
512 int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts); |
8086
ae146e429003
Correct motion vector scaling in B-frames for RV3/4
kostya
parents:
8074
diff
changeset
|
513 int dist = dir ? -GET_PTS_DIFF(r->next_pts, r->cur_pts) : GET_PTS_DIFF(r->cur_pts, r->last_pts); |
ae146e429003
Correct motion vector scaling in B-frames for RV3/4
kostya
parents:
8074
diff
changeset
|
514 int mul; |
6096
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
515 |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
516 if(!refdist) return 0; |
8086
ae146e429003
Correct motion vector scaling in B-frames for RV3/4
kostya
parents:
8074
diff
changeset
|
517 mul = (dist << 14) / refdist; |
ae146e429003
Correct motion vector scaling in B-frames for RV3/4
kostya
parents:
8074
diff
changeset
|
518 return (val * mul + 0x2000) >> 14; |
6096
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
519 } |
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
520 |
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
521 /** |
6026 | 522 * Predict motion vector for B-frame macroblock. |
523 */ | |
524 static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2], | |
525 int A_avail, int B_avail, int C_avail, | |
526 int *mx, int *my) | |
527 { | |
528 if(A_avail + B_avail + C_avail != 3){ | |
529 *mx = A[0] + B[0] + C[0]; | |
530 *my = A[1] + B[1] + C[1]; | |
531 if(A_avail + B_avail + C_avail == 2){ | |
532 *mx /= 2; | |
533 *my /= 2; | |
534 } | |
535 }else{ | |
536 *mx = mid_pred(A[0], B[0], C[0]); | |
537 *my = mid_pred(A[1], B[1], C[1]); | |
538 } | |
539 } | |
540 | |
541 /** | |
542 * motion vector prediction for B-frames | |
543 */ | |
544 static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir) | |
545 { | |
546 MpegEncContext *s = &r->s; | |
547 int mb_pos = s->mb_x + s->mb_y * s->mb_stride; | |
548 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; | |
549 int A[2], B[2], C[2]; | |
550 int has_A = 0, has_B = 0, has_C = 0; | |
551 int mx, my; | |
552 int i, j; | |
553 Picture *cur_pic = s->current_picture_ptr; | |
554 const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0; | |
555 int type = cur_pic->mb_type[mb_pos]; | |
556 | |
557 memset(A, 0, sizeof(A)); | |
558 memset(B, 0, sizeof(B)); | |
559 memset(C, 0, sizeof(C)); | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
560 if((r->avail_cache[6-1] & type) & mask){ |
6026 | 561 A[0] = cur_pic->motion_val[dir][mv_pos - 1][0]; |
562 A[1] = cur_pic->motion_val[dir][mv_pos - 1][1]; | |
563 has_A = 1; | |
564 } | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
565 if((r->avail_cache[6-4] & type) & mask){ |
6026 | 566 B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0]; |
567 B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1]; | |
568 has_B = 1; | |
569 } | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
570 if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){ |
6026 | 571 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0]; |
572 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1]; | |
573 has_C = 1; | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
574 }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){ |
6026 | 575 C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0]; |
576 C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1]; | |
577 has_C = 1; | |
578 } | |
579 | |
580 rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my); | |
581 | |
582 mx += r->dmv[dir][0]; | |
583 my += r->dmv[dir][1]; | |
6096
89140b93ae09
Direct blocks should use motion vectors from the second reference frame
kostya
parents:
6036
diff
changeset
|
584 |
6026 | 585 for(j = 0; j < 2; j++){ |
586 for(i = 0; i < 2; i++){ | |
587 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx; | |
588 cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my; | |
589 } | |
590 } | |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
591 if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){ |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
592 ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride); |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
593 } |
6026 | 594 } |
595 | |
8099 | 596 /** |
597 * motion vector prediction - RV3 version | |
598 */ | |
599 static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir) | |
600 { | |
601 MpegEncContext *s = &r->s; | |
602 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; | |
603 int A[2] = {0}, B[2], C[2]; | |
8504
3bfb7a2ea222
It turned out that RV30 uses motion vectors for forward motion B-frame
kostya
parents:
8445
diff
changeset
|
604 int i, j, k; |
8099 | 605 int mx, my; |
606 int avail_index = avail_indexes[0]; | |
607 | |
608 if(r->avail_cache[avail_index - 1]){ | |
609 A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0]; | |
610 A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1]; | |
611 } | |
612 if(r->avail_cache[avail_index - 4]){ | |
613 B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0]; | |
614 B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1]; | |
615 }else{ | |
616 B[0] = A[0]; | |
617 B[1] = A[1]; | |
618 } | |
619 if(!r->avail_cache[avail_index - 4 + 2]){ | |
620 if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){ | |
621 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0]; | |
622 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1]; | |
623 }else{ | |
624 C[0] = A[0]; | |
625 C[1] = A[1]; | |
626 } | |
627 }else{ | |
628 C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][0]; | |
629 C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+2][1]; | |
630 } | |
631 mx = mid_pred(A[0], B[0], C[0]); | |
632 my = mid_pred(A[1], B[1], C[1]); | |
633 mx += r->dmv[0][0]; | |
634 my += r->dmv[0][1]; | |
635 for(j = 0; j < 2; j++){ | |
636 for(i = 0; i < 2; i++){ | |
8504
3bfb7a2ea222
It turned out that RV30 uses motion vectors for forward motion B-frame
kostya
parents:
8445
diff
changeset
|
637 for(k = 0; k < 2; k++){ |
3bfb7a2ea222
It turned out that RV30 uses motion vectors for forward motion B-frame
kostya
parents:
8445
diff
changeset
|
638 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx; |
3bfb7a2ea222
It turned out that RV30 uses motion vectors for forward motion B-frame
kostya
parents:
8445
diff
changeset
|
639 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my; |
3bfb7a2ea222
It turned out that RV30 uses motion vectors for forward motion B-frame
kostya
parents:
8445
diff
changeset
|
640 } |
8099 | 641 } |
642 } | |
643 } | |
644 | |
8074 | 645 static const int chroma_coeffs[3] = { 0, 3, 5 }; |
6106 | 646 |
6026 | 647 /** |
648 * generic motion compensation function | |
649 * | |
650 * @param r decoder context | |
651 * @param block_type type of the current block | |
652 * @param xoff horizontal offset from the start of the current block | |
653 * @param yoff vertical offset from the start of the current block | |
654 * @param mv_off offset to the motion vector information | |
655 * @param width width of the current partition in 8x8 blocks | |
656 * @param height height of the current partition in 8x8 blocks | |
8684 | 657 * @param dir motion compensation direction (i.e. from the last or the next reference frame) |
658 * @param thirdpel motion vectors are specified in 1/3 of pixel | |
659 * @param qpel_mc a set of functions used to perform luma motion compensation | |
660 * @param chroma_mc a set of functions used to perform chroma motion compensation | |
6026 | 661 */ |
662 static inline void rv34_mc(RV34DecContext *r, const int block_type, | |
663 const int xoff, const int yoff, int mv_off, | |
664 const int width, const int height, int dir, | |
665 const int thirdpel, | |
666 qpel_mc_func (*qpel_mc)[16], | |
667 h264_chroma_mc_func (*chroma_mc)) | |
668 { | |
669 MpegEncContext *s = &r->s; | |
670 uint8_t *Y, *U, *V, *srcY, *srcU, *srcV; | |
8074 | 671 int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y; |
6026 | 672 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off; |
673 int is16x16 = 1; | |
674 | |
675 if(thirdpel){ | |
8074 | 676 int chroma_mx, chroma_my; |
6106 | 677 mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24); |
678 my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24); | |
679 lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3; | |
680 ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3; | |
8074 | 681 chroma_mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + 1) >> 1; |
682 chroma_my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + 1) >> 1; | |
683 umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24); | |
684 umy = (chroma_my + (3 << 24)) / 3 - (1 << 24); | |
685 uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3]; | |
686 uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3]; | |
6026 | 687 }else{ |
8221 | 688 int cx, cy; |
6026 | 689 mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2; |
690 my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2; | |
6121
bc59962f70b9
Fractional parts of motion vectors should be accounted separately too
kostya
parents:
6106
diff
changeset
|
691 lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3; |
bc59962f70b9
Fractional parts of motion vectors should be accounted separately too
kostya
parents:
6106
diff
changeset
|
692 ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3; |
8221 | 693 cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2; |
694 cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2; | |
695 umx = cx >> 2; | |
696 umy = cy >> 2; | |
697 uvmx = (cx & 3) << 1; | |
698 uvmy = (cy & 3) << 1; | |
699 //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3 | |
700 if(uvmx == 6 && uvmy == 6) | |
701 uvmx = uvmy = 4; | |
6026 | 702 } |
6121
bc59962f70b9
Fractional parts of motion vectors should be accounted separately too
kostya
parents:
6106
diff
changeset
|
703 dxy = ly*4 + lx; |
6026 | 704 srcY = dir ? s->next_picture_ptr->data[0] : s->last_picture_ptr->data[0]; |
705 srcU = dir ? s->next_picture_ptr->data[1] : s->last_picture_ptr->data[1]; | |
706 srcV = dir ? s->next_picture_ptr->data[2] : s->last_picture_ptr->data[2]; | |
707 src_x = s->mb_x * 16 + xoff + mx; | |
708 src_y = s->mb_y * 16 + yoff + my; | |
8074 | 709 uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx; |
710 uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy; | |
6026 | 711 srcY += src_y * s->linesize + src_x; |
712 srcU += uvsrc_y * s->uvlinesize + uvsrc_x; | |
713 srcV += uvsrc_y * s->uvlinesize + uvsrc_x; | |
8179 | 714 if( (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 |
715 || (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4){ | |
716 uint8_t *uvbuf= s->edge_emu_buffer + 22 * s->linesize; | |
6026 | 717 |
718 srcY -= 2 + 2*s->linesize; | |
8179 | 719 ff_emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6, |
6026 | 720 src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos); |
721 srcY = s->edge_emu_buffer + 2 + 2*s->linesize; | |
722 ff_emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1, | |
723 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); | |
724 ff_emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1, | |
725 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); | |
726 srcU = uvbuf; | |
727 srcV = uvbuf + 16; | |
728 } | |
729 Y = s->dest[0] + xoff + yoff *s->linesize; | |
730 U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize; | |
731 V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize; | |
732 | |
733 if(block_type == RV34_MB_P_16x8){ | |
734 qpel_mc[1][dxy](Y, srcY, s->linesize); | |
735 Y += 8; | |
736 srcY += 8; | |
737 }else if(block_type == RV34_MB_P_8x16){ | |
738 qpel_mc[1][dxy](Y, srcY, s->linesize); | |
739 Y += 8 * s->linesize; | |
740 srcY += 8 * s->linesize; | |
741 } | |
742 is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16); | |
743 qpel_mc[!is16x16][dxy](Y, srcY, s->linesize); | |
744 chroma_mc[2-width] (U, srcU, s->uvlinesize, height*4, uvmx, uvmy); | |
745 chroma_mc[2-width] (V, srcV, s->uvlinesize, height*4, uvmx, uvmy); | |
746 } | |
747 | |
748 static void rv34_mc_1mv(RV34DecContext *r, const int block_type, | |
749 const int xoff, const int yoff, int mv_off, | |
750 const int width, const int height, int dir) | |
751 { | |
752 rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, | |
6106 | 753 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab |
8221 | 754 : r->s.dsp.put_rv40_qpel_pixels_tab, |
755 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab | |
756 : r->s.dsp.put_rv40_chroma_pixels_tab); | |
6026 | 757 } |
758 | |
759 static void rv34_mc_2mv(RV34DecContext *r, const int block_type) | |
760 { | |
761 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, | |
6106 | 762 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab |
8221 | 763 : r->s.dsp.put_rv40_qpel_pixels_tab, |
8244
3f3d653fb46d
34l: forgot to change one case of chroma MC for RV40
kostya
parents:
8243
diff
changeset
|
764 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab |
3f3d653fb46d
34l: forgot to change one case of chroma MC for RV40
kostya
parents:
8243
diff
changeset
|
765 : r->s.dsp.put_rv40_chroma_pixels_tab); |
6026 | 766 rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, |
6106 | 767 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab |
8221 | 768 : r->s.dsp.avg_rv40_qpel_pixels_tab, |
769 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab | |
770 : r->s.dsp.avg_rv40_chroma_pixels_tab); | |
6026 | 771 } |
772 | |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
773 static void rv34_mc_2mv_skip(RV34DecContext *r) |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
774 { |
8128 | 775 int i, j; |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
776 for(j = 0; j < 2; j++) |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
777 for(i = 0; i < 2; i++){ |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
778 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30, |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
779 r->rv30 ? r->s.dsp.put_rv30_tpel_pixels_tab |
8221 | 780 : r->s.dsp.put_rv40_qpel_pixels_tab, |
781 r->rv30 ? r->s.dsp.put_h264_chroma_pixels_tab | |
782 : r->s.dsp.put_rv40_chroma_pixels_tab); | |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
783 rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30, |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
784 r->rv30 ? r->s.dsp.avg_rv30_tpel_pixels_tab |
8221 | 785 : r->s.dsp.avg_rv40_qpel_pixels_tab, |
786 r->rv30 ? r->s.dsp.avg_h264_chroma_pixels_tab | |
787 : r->s.dsp.avg_rv40_chroma_pixels_tab); | |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
788 } |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
789 } |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
790 |
6026 | 791 /** number of motion vectors in each macroblock type */ |
792 static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 }; | |
793 | |
794 /** | |
795 * Decode motion vector differences | |
796 * and perform motion vector reconstruction and motion compensation. | |
797 */ | |
798 static int rv34_decode_mv(RV34DecContext *r, int block_type) | |
799 { | |
800 MpegEncContext *s = &r->s; | |
801 GetBitContext *gb = &s->gb; | |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
802 int i, j, k, l; |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
803 int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
804 int next_bt; |
6026 | 805 |
806 memset(r->dmv, 0, sizeof(r->dmv)); | |
807 for(i = 0; i < num_mvs[block_type]; i++){ | |
808 r->dmv[i][0] = svq3_get_se_golomb(gb); | |
809 r->dmv[i][1] = svq3_get_se_golomb(gb); | |
810 } | |
811 switch(block_type){ | |
812 case RV34_MB_TYPE_INTRA: | |
813 case RV34_MB_TYPE_INTRA16x16: | |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
814 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); |
6026 | 815 return 0; |
816 case RV34_MB_SKIP: | |
6481 | 817 if(s->pict_type == FF_P_TYPE){ |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
818 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); |
6026 | 819 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0); |
820 break; | |
821 } | |
822 case RV34_MB_B_DIRECT: | |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
823 //surprisingly, it uses motion scheme from next reference frame |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
824 next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride]; |
9162 | 825 if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){ |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
826 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); |
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
827 ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); |
9162 | 828 }else |
9041 | 829 for(j = 0; j < 2; j++) |
830 for(i = 0; i < 2; i++) | |
831 for(k = 0; k < 2; k++) | |
832 for(l = 0; l < 2; l++) | |
833 s->current_picture_ptr->motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][k]); | |
9162 | 834 if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC |
6693
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
835 rv34_mc_2mv(r, block_type); |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
836 else |
6f13852a9161
Skip blocks in B-frames reuse motion vectors from next reference frame.
kostya
parents:
6517
diff
changeset
|
837 rv34_mc_2mv_skip(r); |
10118
4dafbd9c8918
Zeroing pic->motion_val in RV3/4 causes alignment problems on some 64-bit
kostya
parents:
10117
diff
changeset
|
838 ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride); |
6026 | 839 break; |
840 case RV34_MB_P_16x16: | |
841 case RV34_MB_P_MIX16x16: | |
842 rv34_pred_mv(r, block_type, 0, 0); | |
843 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0); | |
844 break; | |
845 case RV34_MB_B_FORWARD: | |
846 case RV34_MB_B_BACKWARD: | |
847 r->dmv[1][0] = r->dmv[0][0]; | |
848 r->dmv[1][1] = r->dmv[0][1]; | |
8099 | 849 if(r->rv30) |
850 rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD); | |
851 else | |
852 rv34_pred_mv_b (r, block_type, block_type == RV34_MB_B_BACKWARD); | |
6026 | 853 rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD); |
854 break; | |
855 case RV34_MB_P_16x8: | |
856 case RV34_MB_P_8x16: | |
857 rv34_pred_mv(r, block_type, 0, 0); | |
858 rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1); | |
859 if(block_type == RV34_MB_P_16x8){ | |
860 rv34_mc_1mv(r, block_type, 0, 0, 0, 2, 1, 0); | |
861 rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0); | |
862 } | |
863 if(block_type == RV34_MB_P_8x16){ | |
864 rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0); | |
865 rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0); | |
866 } | |
867 break; | |
868 case RV34_MB_B_BIDIR: | |
869 rv34_pred_mv_b (r, block_type, 0); | |
870 rv34_pred_mv_b (r, block_type, 1); | |
871 rv34_mc_2mv (r, block_type); | |
872 break; | |
873 case RV34_MB_P_8x8: | |
874 for(i=0;i< 4;i++){ | |
875 rv34_pred_mv(r, block_type, i, i); | |
876 rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0); | |
877 } | |
878 break; | |
879 } | |
880 | |
881 return 0; | |
882 } | |
883 /** @} */ // mv group | |
884 | |
885 /** | |
886 * @defgroup recons Macroblock reconstruction functions | |
887 * @{ | |
888 */ | |
889 /** mapping of RV30/40 intra prediction types to standard H.264 types */ | |
890 static const int ittrans[9] = { | |
891 DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED, | |
892 VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED, | |
893 }; | |
894 | |
895 /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */ | |
896 static const int ittrans16[4] = { | |
897 DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8, | |
898 }; | |
899 | |
900 /** | |
901 * Perform 4x4 intra prediction. | |
902 */ | |
903 static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right) | |
904 { | |
905 uint8_t *prev = dst - stride + 4; | |
906 uint32_t topleft; | |
907 | |
908 if(!up && !left) | |
909 itype = DC_128_PRED; | |
910 else if(!up){ | |
911 if(itype == VERT_PRED) itype = HOR_PRED; | |
912 if(itype == DC_PRED) itype = LEFT_DC_PRED; | |
913 }else if(!left){ | |
914 if(itype == HOR_PRED) itype = VERT_PRED; | |
915 if(itype == DC_PRED) itype = TOP_DC_PRED; | |
916 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN; | |
917 } | |
918 if(!down){ | |
919 if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN; | |
920 if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN; | |
6036
ce3b68242317
Correct spatial prediction mode in RV30/40 for vertical left direction
kostya
parents:
6026
diff
changeset
|
921 if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN; |
6026 | 922 } |
923 if(!right && up){ | |
924 topleft = dst[-stride + 3] * 0x01010101; | |
8721
42e258f2b931
Silence two pointer assignment compiler warnings in rv34.c
kostya
parents:
8718
diff
changeset
|
925 prev = (uint8_t*)&topleft; |
6026 | 926 } |
927 r->h.pred4x4[itype](dst, prev, stride); | |
928 } | |
929 | |
930 /** add_pixels_clamped for 4x4 block */ | |
931 static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off) | |
932 { | |
933 int x, y; | |
934 for(y = 0; y < 4; y++) | |
935 for(x = 0; x < 4; x++) | |
936 dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]); | |
937 } | |
938 | |
939 static inline int adjust_pred16(int itype, int up, int left) | |
940 { | |
941 if(!up && !left) | |
942 itype = DC_128_PRED8x8; | |
943 else if(!up){ | |
944 if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8; | |
945 if(itype == VERT_PRED8x8) itype = HOR_PRED8x8; | |
946 if(itype == DC_PRED8x8) itype = LEFT_DC_PRED8x8; | |
947 }else if(!left){ | |
948 if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8; | |
949 if(itype == HOR_PRED8x8) itype = VERT_PRED8x8; | |
950 if(itype == DC_PRED8x8) itype = TOP_DC_PRED8x8; | |
951 } | |
952 return itype; | |
953 } | |
954 | |
955 static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16) | |
956 { | |
957 MpegEncContext *s = &r->s; | |
958 DSPContext *dsp = &s->dsp; | |
959 int i, j; | |
960 uint8_t *Y, *U, *V; | |
961 int itype; | |
962 int avail[6*8] = {0}; | |
963 int idx; | |
964 | |
965 // Set neighbour information. | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
966 if(r->avail_cache[1]) |
6026 | 967 avail[0] = 1; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
968 if(r->avail_cache[2]) |
6026 | 969 avail[1] = avail[2] = 1; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
970 if(r->avail_cache[3]) |
6026 | 971 avail[3] = avail[4] = 1; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
972 if(r->avail_cache[4]) |
6026 | 973 avail[5] = 1; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
974 if(r->avail_cache[5]) |
6026 | 975 avail[8] = avail[16] = 1; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
976 if(r->avail_cache[9]) |
6026 | 977 avail[24] = avail[32] = 1; |
978 | |
979 Y = s->dest[0]; | |
980 U = s->dest[1]; | |
981 V = s->dest[2]; | |
982 if(!is16){ | |
983 for(j = 0; j < 4; j++){ | |
984 idx = 9 + j*8; | |
985 for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){ | |
986 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]); | |
987 avail[idx] = 1; | |
988 if(cbp & 1) | |
989 rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32); | |
990 } | |
991 Y += s->linesize * 4 - 4*4; | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
992 intra_types += r->intra_types_stride; |
6026 | 993 } |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
994 intra_types -= r->intra_types_stride * 4; |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
995 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4); |
6026 | 996 for(j = 0; j < 2; j++){ |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
997 idx = 6 + j*4; |
6026 | 998 for(i = 0; i < 2; i++, cbp >>= 1, idx++){ |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
999 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]); |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1000 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]); |
6026 | 1001 r->avail_cache[idx] = 1; |
1002 if(cbp & 0x01) | |
1003 rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32); | |
1004 if(cbp & 0x10) | |
1005 rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32); | |
1006 } | |
1007 } | |
1008 }else{ | |
1009 itype = ittrans16[intra_types[0]]; | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1010 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]); |
6026 | 1011 r->h.pred16x16[itype](Y, s->linesize); |
8180
f4a291734ad7
Eliminate direct use of s->current_picture in RV30/40 decoder
kostya
parents:
8179
diff
changeset
|
1012 dsp->add_pixels_clamped(s->block[0], Y, s->linesize); |
f4a291734ad7
Eliminate direct use of s->current_picture in RV30/40 decoder
kostya
parents:
8179
diff
changeset
|
1013 dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize); |
f4a291734ad7
Eliminate direct use of s->current_picture in RV30/40 decoder
kostya
parents:
8179
diff
changeset
|
1014 Y += s->linesize * 8; |
f4a291734ad7
Eliminate direct use of s->current_picture in RV30/40 decoder
kostya
parents:
8179
diff
changeset
|
1015 dsp->add_pixels_clamped(s->block[2], Y, s->linesize); |
f4a291734ad7
Eliminate direct use of s->current_picture in RV30/40 decoder
kostya
parents:
8179
diff
changeset
|
1016 dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize); |
6026 | 1017 |
1018 itype = ittrans16[intra_types[0]]; | |
1019 if(itype == PLANE_PRED8x8) itype = DC_PRED8x8; | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1020 itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]); |
6026 | 1021 r->h.pred8x8[itype](U, s->uvlinesize); |
1022 dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize); | |
1023 r->h.pred8x8[itype](V, s->uvlinesize); | |
1024 dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize); | |
1025 } | |
1026 } | |
1027 | |
1028 /** @} */ // recons group | |
1029 | |
1030 /** | |
1031 * @addtogroup bitstream | |
1032 * Decode macroblock header and return CBP in case of success, -1 otherwise. | |
1033 */ | |
1034 static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types) | |
1035 { | |
1036 MpegEncContext *s = &r->s; | |
1037 GetBitContext *gb = &s->gb; | |
1038 int mb_pos = s->mb_x + s->mb_y * s->mb_stride; | |
1039 int i, t; | |
1040 | |
1041 if(!r->si.type){ | |
1042 r->is16 = get_bits1(gb); | |
1043 if(!r->is16 && !r->rv30){ | |
1044 if(!get_bits1(gb)) | |
1045 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n"); | |
1046 } | |
1047 s->current_picture_ptr->mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA; | |
1048 r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA; | |
1049 }else{ | |
1050 r->block_type = r->decode_mb_info(r); | |
1051 if(r->block_type == -1) | |
1052 return -1; | |
1053 s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type]; | |
1054 r->mb_type[mb_pos] = r->block_type; | |
1055 if(r->block_type == RV34_MB_SKIP){ | |
6481 | 1056 if(s->pict_type == FF_P_TYPE) |
6026 | 1057 r->mb_type[mb_pos] = RV34_MB_P_16x16; |
6481 | 1058 if(s->pict_type == FF_B_TYPE) |
6026 | 1059 r->mb_type[mb_pos] = RV34_MB_B_DIRECT; |
1060 } | |
1061 r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]); | |
1062 rv34_decode_mv(r, r->block_type); | |
1063 if(r->block_type == RV34_MB_SKIP){ | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1064 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0])); |
6026 | 1065 return 0; |
1066 } | |
1067 r->chroma_vlc = 1; | |
1068 r->luma_vlc = 0; | |
1069 } | |
1070 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){ | |
1071 if(r->is16){ | |
1072 t = get_bits(gb, 2); | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1073 fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0])); |
6026 | 1074 r->luma_vlc = 2; |
1075 }else{ | |
1076 if(r->decode_intra_types(r, gb, intra_types) < 0) | |
1077 return -1; | |
1078 r->luma_vlc = 1; | |
1079 } | |
1080 r->chroma_vlc = 0; | |
1081 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0); | |
1082 }else{ | |
1083 for(i = 0; i < 16; i++) | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1084 intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0; |
6026 | 1085 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1); |
1086 if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){ | |
1087 r->is16 = 1; | |
1088 r->chroma_vlc = 1; | |
1089 r->luma_vlc = 2; | |
1090 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0); | |
1091 } | |
1092 } | |
1093 | |
1094 return rv34_decode_cbp(gb, r->cur_vlcs, r->is16); | |
1095 } | |
1096 | |
1097 /** | |
1098 * @addtogroup recons | |
1099 * @{ | |
1100 */ | |
1101 /** | |
1102 * mask for retrieving all bits in coded block pattern | |
1103 * corresponding to one 8x8 block | |
1104 */ | |
8181
d8f516ca0002
33l Fix applying residue condition in RV30/40 decoder
kostya
parents:
8180
diff
changeset
|
1105 #define LUMA_CBP_BLOCK_MASK 0x33 |
6026 | 1106 |
1107 #define U_CBP_MASK 0x0F0000 | |
1108 #define V_CBP_MASK 0xF00000 | |
1109 | |
1110 | |
1111 static void rv34_apply_differences(RV34DecContext *r, int cbp) | |
1112 { | |
1113 static const int shifts[4] = { 0, 2, 8, 10 }; | |
1114 MpegEncContext *s = &r->s; | |
1115 int i; | |
1116 | |
1117 for(i = 0; i < 4; i++) | |
8181
d8f516ca0002
33l Fix applying residue condition in RV30/40 decoder
kostya
parents:
8180
diff
changeset
|
1118 if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16) |
6026 | 1119 s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize); |
1120 if(cbp & U_CBP_MASK) | |
1121 s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize); | |
1122 if(cbp & V_CBP_MASK) | |
1123 s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize); | |
1124 } | |
1125 | |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1126 static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step) |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1127 { |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1128 int d; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1129 d = motion_val[0][0] - motion_val[-step][0]; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1130 if(d < -3 || d > 3) |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1131 return 1; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1132 d = motion_val[0][1] - motion_val[-step][1]; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1133 if(d < -3 || d > 3) |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1134 return 1; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1135 return 0; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1136 } |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1137 |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1138 static int rv34_set_deblock_coef(RV34DecContext *r) |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1139 { |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1140 MpegEncContext *s = &r->s; |
8122
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1141 int hmvmask = 0, vmvmask = 0, i, j; |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1142 int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1143 int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx]; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1144 for(j = 0; j < 16; j += 8){ |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1145 for(i = 0; i < 2; i++){ |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1146 if(is_mv_diff_gt_3(motion_val + i, 1)) |
8122
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1147 vmvmask |= 0x11 << (j + i*2); |
8245
7e6ca1be9e40
Fix reading out of buffer during RV30/40 deblock mask calculation
kostya
parents:
8244
diff
changeset
|
1148 if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride)) |
8122
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1149 hmvmask |= 0x03 << (j + i*2); |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1150 } |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1151 motion_val += s->b8_stride; |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1152 } |
8122
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1153 if(s->first_slice_line) |
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1154 hmvmask &= ~0x000F; |
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1155 if(!s->mb_x) |
aec296246352
Split RV3/4 deblock pattern into horizontal and vertical parts
kostya
parents:
8121
diff
changeset
|
1156 vmvmask &= ~0x1111; |
8368 | 1157 if(r->rv30){ //RV30 marks both subblocks on the edge for filtering |
1158 vmvmask |= (vmvmask & 0x4444) >> 1; | |
1159 hmvmask |= (hmvmask & 0x0F00) >> 4; | |
1160 if(s->mb_x) | |
1161 r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3; | |
1162 if(!s->first_slice_line) | |
1163 r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12; | |
1164 } | |
1165 return hmvmask | vmvmask; | |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1166 } |
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1167 |
6026 | 1168 static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types) |
1169 { | |
1170 MpegEncContext *s = &r->s; | |
1171 GetBitContext *gb = &s->gb; | |
1172 int cbp, cbp2; | |
1173 int i, blknum, blkoff; | |
1174 DCTELEM block16[64]; | |
1175 int luma_dc_quant; | |
1176 int dist; | |
1177 int mb_pos = s->mb_x + s->mb_y * s->mb_stride; | |
1178 | |
1179 // Calculate which neighbours are available. Maybe it's worth optimizing too. | |
1180 memset(r->avail_cache, 0, sizeof(r->avail_cache)); | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1181 fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4); |
6026 | 1182 dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width; |
1183 if(s->mb_x && dist) | |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1184 r->avail_cache[5] = |
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1185 r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1]; |
6026 | 1186 if(dist >= s->mb_width) |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1187 r->avail_cache[2] = |
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1188 r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride]; |
6026 | 1189 if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1) |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1190 r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1]; |
6026 | 1191 if(s->mb_x && dist > s->mb_width) |
10117
d0f2f4e4436c
Adjust r->avail_cache[] indexes to avoid alignment issues when zeroing block
kostya
parents:
9903
diff
changeset
|
1192 r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1]; |
6026 | 1193 |
1194 s->qscale = r->si.quant; | |
1195 cbp = cbp2 = rv34_decode_mb_header(r, intra_types); | |
8370 | 1196 r->cbp_luma [mb_pos] = cbp; |
1197 r->cbp_chroma[mb_pos] = cbp >> 16; | |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1198 if(s->pict_type == FF_I_TYPE) |
8369
45c3780b1a96
Merge deblocking pattern with CBP for RV30/40 loop filtering
kostya
parents:
8368
diff
changeset
|
1199 r->deblock_coefs[mb_pos] = 0xFFFF; |
8116
2d01559f824c
Calculating an additional MV-based deblocking pattern is the same
kostya
parents:
8099
diff
changeset
|
1200 else |
8369
45c3780b1a96
Merge deblocking pattern with CBP for RV30/40 loop filtering
kostya
parents:
8368
diff
changeset
|
1201 r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos]; |
8370 | 1202 s->current_picture_ptr->qscale_table[mb_pos] = s->qscale; |
6026 | 1203 |
1204 if(cbp == -1) | |
1205 return -1; | |
1206 | |
8046 | 1207 luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale]; |
6026 | 1208 if(r->is16){ |
1209 memset(block16, 0, sizeof(block16)); | |
1210 rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0); | |
1211 rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]); | |
1212 rv34_inv_transform_noround(block16); | |
1213 } | |
1214 | |
1215 for(i = 0; i < 16; i++, cbp >>= 1){ | |
1216 if(!r->is16 && !(cbp & 1)) continue; | |
1217 blknum = ((i & 2) >> 1) + ((i & 8) >> 2); | |
1218 blkoff = ((i & 1) << 2) + ((i & 4) << 3); | |
1219 if(cbp & 1) | |
1220 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0); | |
8046 | 1221 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]); |
6026 | 1222 if(r->is16) //FIXME: optimize |
1223 s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)]; | |
1224 rv34_inv_transform(s->block[blknum] + blkoff); | |
1225 } | |
1226 if(r->block_type == RV34_MB_P_MIX16x16) | |
1227 r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1); | |
1228 for(; i < 24; i++, cbp >>= 1){ | |
1229 if(!(cbp & 1)) continue; | |
1230 blknum = ((i & 4) >> 2) + 4; | |
1231 blkoff = ((i & 1) << 2) + ((i & 2) << 4); | |
1232 rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1); | |
1233 rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]); | |
1234 rv34_inv_transform(s->block[blknum] + blkoff); | |
1235 } | |
8370 | 1236 if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])) |
6026 | 1237 rv34_output_macroblock(r, intra_types, cbp2, r->is16); |
1238 else | |
1239 rv34_apply_differences(r, cbp2); | |
1240 | |
1241 return 0; | |
1242 } | |
1243 | |
1244 static int check_slice_end(RV34DecContext *r, MpegEncContext *s) | |
1245 { | |
1246 int bits; | |
1247 if(s->mb_y >= s->mb_height) | |
1248 return 1; | |
1249 if(!s->mb_num_left) | |
1250 return 1; | |
1251 if(r->s.mb_skip_run > 1) | |
1252 return 0; | |
1253 bits = r->bits - get_bits_count(&s->gb); | |
1254 if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits))) | |
1255 return 1; | |
1256 return 0; | |
1257 } | |
1258 | |
1259 static inline int slice_compare(SliceInfo *si1, SliceInfo *si2) | |
1260 { | |
1261 return si1->type != si2->type || | |
1262 si1->start >= si2->start || | |
1263 si1->width != si2->width || | |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1264 si1->height != si2->height|| |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1265 si1->pts != si2->pts; |
6026 | 1266 } |
1267 | |
8371
2ebc9b2c7459
add const qualifier to some pointers for input data
kostya
parents:
8370
diff
changeset
|
1268 static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size) |
6026 | 1269 { |
1270 MpegEncContext *s = &r->s; | |
1271 GetBitContext *gb = &s->gb; | |
1272 int mb_pos; | |
1273 int res; | |
1274 | |
1275 init_get_bits(&r->s.gb, buf, buf_size*8); | |
1276 res = r->parse_slice_header(r, gb, &r->si); | |
1277 if(res < 0){ | |
1278 av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n"); | |
1279 return -1; | |
1280 } | |
1281 | |
1282 if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) { | |
1283 if(s->width != r->si.width || s->height != r->si.height){ | |
1284 av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height); | |
1285 MPV_common_end(s); | |
1286 s->width = r->si.width; | |
1287 s->height = r->si.height; | |
1288 if(MPV_common_init(s) < 0) | |
1289 return -1; | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1290 r->intra_types_stride = s->mb_width*4 + 4; |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1291 r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist)); |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1292 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4; |
6026 | 1293 r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type)); |
6155
a425bdc70ac5
Save coded block patterns for future loop filtering.
kostya
parents:
6121
diff
changeset
|
1294 r->cbp_luma = av_realloc(r->cbp_luma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma)); |
a425bdc70ac5
Save coded block patterns for future loop filtering.
kostya
parents:
6121
diff
changeset
|
1295 r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma)); |
8037
b6d93bef844a
RV3 and RV4 decoders set some deblocking coefs for each macroblock,
kostya
parents:
8036
diff
changeset
|
1296 r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs)); |
6026 | 1297 } |
6481 | 1298 s->pict_type = r->si.type ? r->si.type : FF_I_TYPE; |
6026 | 1299 if(MPV_frame_start(s, s->avctx) < 0) |
1300 return -1; | |
1301 ff_er_frame_start(s); | |
6714
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1302 r->cur_pts = r->si.pts; |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1303 if(s->pict_type != FF_B_TYPE){ |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1304 r->last_pts = r->next_pts; |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1305 r->next_pts = r->cur_pts; |
05c3a4b419e9
Calculate motion vector information based on PTS provided in slice header
kostya
parents:
6693
diff
changeset
|
1306 } |
6026 | 1307 s->mb_x = s->mb_y = 0; |
1308 } | |
1309 | |
1310 r->si.end = end; | |
1311 s->qscale = r->si.quant; | |
1312 r->bits = buf_size*8; | |
1313 s->mb_num_left = r->si.end - r->si.start; | |
1314 r->s.mb_skip_run = 0; | |
1315 | |
1316 mb_pos = s->mb_x + s->mb_y * s->mb_width; | |
1317 if(r->si.start != mb_pos){ | |
1318 av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos); | |
1319 s->mb_x = r->si.start % s->mb_width; | |
1320 s->mb_y = r->si.start / s->mb_width; | |
1321 } | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1322 memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist)); |
6026 | 1323 s->first_slice_line = 1; |
1324 s->resync_mb_x= s->mb_x; | |
1325 s->resync_mb_y= s->mb_y; | |
1326 | |
1327 ff_init_block_index(s); | |
1328 while(!check_slice_end(r, s)) { | |
1329 ff_update_block_index(s); | |
1330 s->dsp.clear_blocks(s->block[0]); | |
1331 | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1332 if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){ |
6026 | 1333 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR); |
1334 return -1; | |
1335 } | |
1336 if (++s->mb_x == s->mb_width) { | |
1337 s->mb_x = 0; | |
1338 s->mb_y++; | |
1339 ff_init_block_index(s); | |
1340 | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1341 memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist)); |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1342 memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist)); |
8224
6771b3544991
Invoke future RV30/40 loop filter for already decoded rows instead of
kostya
parents:
8221
diff
changeset
|
1343 |
6771b3544991
Invoke future RV30/40 loop filter for already decoded rows instead of
kostya
parents:
8221
diff
changeset
|
1344 if(r->loop_filter && s->mb_y >= 2) |
6771b3544991
Invoke future RV30/40 loop filter for already decoded rows instead of
kostya
parents:
8221
diff
changeset
|
1345 r->loop_filter(r, s->mb_y - 2); |
6026 | 1346 } |
1347 if(s->mb_x == s->resync_mb_x) | |
1348 s->first_slice_line=0; | |
1349 s->mb_num_left--; | |
1350 } | |
1351 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END); | |
1352 | |
6750 | 1353 return s->mb_y == s->mb_height; |
6026 | 1354 } |
1355 | |
1356 /** @} */ // recons group end | |
1357 | |
1358 /** | |
1359 * Initialize decoder. | |
1360 */ | |
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
6481
diff
changeset
|
1361 av_cold int ff_rv34_decode_init(AVCodecContext *avctx) |
6026 | 1362 { |
1363 RV34DecContext *r = avctx->priv_data; | |
1364 MpegEncContext *s = &r->s; | |
1365 | |
1366 MPV_decode_defaults(s); | |
1367 s->avctx= avctx; | |
1368 s->out_format = FMT_H263; | |
1369 s->codec_id= avctx->codec_id; | |
1370 | |
1371 s->width = avctx->width; | |
1372 s->height = avctx->height; | |
1373 | |
1374 r->s.avctx = avctx; | |
1375 avctx->flags |= CODEC_FLAG_EMU_EDGE; | |
1376 r->s.flags |= CODEC_FLAG_EMU_EDGE; | |
1377 avctx->pix_fmt = PIX_FMT_YUV420P; | |
1378 avctx->has_b_frames = 1; | |
1379 s->low_delay = 0; | |
1380 | |
1381 if (MPV_common_init(s) < 0) | |
1382 return -1; | |
1383 | |
1384 ff_h264_pred_init(&r->h, CODEC_ID_RV40); | |
1385 | |
9903
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1386 r->intra_types_stride = 4*s->mb_stride + 4; |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1387 r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist)); |
305536ce781f
RV3/4 intra types array causes alignment issues (at least on ARM5), thus change
kostya
parents:
9568
diff
changeset
|
1388 r->intra_types = r->intra_types_hist + r->intra_types_stride * 4; |
6026 | 1389 |
1390 r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type)); | |
1391 | |
6155
a425bdc70ac5
Save coded block patterns for future loop filtering.
kostya
parents:
6121
diff
changeset
|
1392 r->cbp_luma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma)); |
a425bdc70ac5
Save coded block patterns for future loop filtering.
kostya
parents:
6121
diff
changeset
|
1393 r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma)); |
8037
b6d93bef844a
RV3 and RV4 decoders set some deblocking coefs for each macroblock,
kostya
parents:
8036
diff
changeset
|
1394 r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs)); |
6155
a425bdc70ac5
Save coded block patterns for future loop filtering.
kostya
parents:
6121
diff
changeset
|
1395 |
9568
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
1396 if(!intra_vlcs[0].cbppattern[0].bits) |
976a7ad342af
Revert changing VLC initialization type for RV3/4 decoder.
kostya
parents:
9554
diff
changeset
|
1397 rv34_init_tables(); |
6026 | 1398 |
1399 return 0; | |
1400 } | |
1401 | |
8371
2ebc9b2c7459
add const qualifier to some pointers for input data
kostya
parents:
8370
diff
changeset
|
1402 static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n) |
6026 | 1403 { |
1404 if(avctx->slice_count) return avctx->slice_offset[n]; | |
1405 else return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) : AV_RB32(buf + n*8); | |
1406 } | |
1407 | |
1408 int ff_rv34_decode_frame(AVCodecContext *avctx, | |
1409 void *data, int *data_size, | |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
9168
diff
changeset
|
1410 AVPacket *avpkt) |
6026 | 1411 { |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
9168
diff
changeset
|
1412 const uint8_t *buf = avpkt->data; |
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
9168
diff
changeset
|
1413 int buf_size = avpkt->size; |
6026 | 1414 RV34DecContext *r = avctx->priv_data; |
1415 MpegEncContext *s = &r->s; | |
1416 AVFrame *pict = data; | |
1417 SliceInfo si; | |
1418 int i; | |
1419 int slice_count; | |
8371
2ebc9b2c7459
add const qualifier to some pointers for input data
kostya
parents:
8370
diff
changeset
|
1420 const uint8_t *slices_hdr = NULL; |
6026 | 1421 int last = 0; |
1422 | |
1423 /* no supplementary picture */ | |
1424 if (buf_size == 0) { | |
1425 /* special case for last picture */ | |
1426 if (s->low_delay==0 && s->next_picture_ptr) { | |
1427 *pict= *(AVFrame*)s->next_picture_ptr; | |
1428 s->next_picture_ptr= NULL; | |
1429 | |
1430 *data_size = sizeof(AVFrame); | |
1431 } | |
1432 return 0; | |
1433 } | |
1434 | |
1435 if(!avctx->slice_count){ | |
1436 slice_count = (*buf++) + 1; | |
1437 slices_hdr = buf + 4; | |
1438 buf += 8 * slice_count; | |
1439 }else | |
1440 slice_count = avctx->slice_count; | |
1441 | |
8754
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1442 //parse first slice header to check whether this frame can be decoded |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1443 if(get_slice_offset(avctx, slices_hdr, 0) > buf_size){ |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1444 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n"); |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1445 return -1; |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1446 } |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1447 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), buf_size-get_slice_offset(avctx, slices_hdr, 0)); |
8821
58706d84c938
In RV3/4 nonzero MB offset for the first slice also indicates error
kostya
parents:
8754
diff
changeset
|
1448 if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){ |
8754
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1449 av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n"); |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1450 return -1; |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1451 } |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1452 if((!s->last_picture_ptr || !s->last_picture_ptr->data[0]) && si.type == FF_B_TYPE) |
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1453 return -1; |
8827 | 1454 /* skip b frames if we are in a hurry */ |
1455 if(avctx->hurry_up && si.type==FF_B_TYPE) return buf_size; | |
1456 if( (avctx->skip_frame >= AVDISCARD_NONREF && si.type==FF_B_TYPE) | |
1457 || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=FF_I_TYPE) | |
1458 || avctx->skip_frame >= AVDISCARD_ALL) | |
1459 return buf_size; | |
1460 /* skip everything if we are in a hurry>=5 */ | |
1461 if(avctx->hurry_up>=5) | |
1462 return buf_size; | |
8754
f471abd09fef
Move RV3/4 frame type check up so 1-slice unanchored B-frames are checked, too.
kostya
parents:
8721
diff
changeset
|
1463 |
6026 | 1464 for(i=0; i<slice_count; i++){ |
1465 int offset= get_slice_offset(avctx, slices_hdr, i); | |
1466 int size; | |
1467 if(i+1 == slice_count) | |
1468 size= buf_size - offset; | |
1469 else | |
1470 size= get_slice_offset(avctx, slices_hdr, i+1) - offset; | |
1471 | |
8243 | 1472 if(offset > buf_size){ |
1473 av_log(avctx, AV_LOG_ERROR, "Slice offset is greater than frame size\n"); | |
1474 break; | |
1475 } | |
1476 | |
6026 | 1477 r->si.end = s->mb_width * s->mb_height; |
1478 if(i+1 < slice_count){ | |
1479 init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8); | |
1480 if(r->parse_slice_header(r, &r->s.gb, &si) < 0){ | |
1481 if(i+2 < slice_count) | |
1482 size = get_slice_offset(avctx, slices_hdr, i+2) - offset; | |
1483 else | |
1484 size = buf_size - offset; | |
1485 }else | |
1486 r->si.end = si.start; | |
1487 } | |
1488 last = rv34_decode_slice(r, r->si.end, buf + offset, size); | |
1489 s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start; | |
1490 if(last) | |
1491 break; | |
1492 } | |
1493 | |
1494 if(last){ | |
1495 if(r->loop_filter) | |
8224
6771b3544991
Invoke future RV30/40 loop filter for already decoded rows instead of
kostya
parents:
8221
diff
changeset
|
1496 r->loop_filter(r, s->mb_height - 1); |
6026 | 1497 ff_er_frame_end(s); |
1498 MPV_frame_end(s); | |
6481 | 1499 if (s->pict_type == FF_B_TYPE || s->low_delay) { |
6026 | 1500 *pict= *(AVFrame*)s->current_picture_ptr; |
1501 } else if (s->last_picture_ptr != NULL) { | |
1502 *pict= *(AVFrame*)s->last_picture_ptr; | |
1503 } | |
1504 | |
1505 if(s->last_picture_ptr || s->low_delay){ | |
1506 *data_size = sizeof(AVFrame); | |
1507 ff_print_debug_info(s, pict); | |
1508 } | |
1509 s->current_picture_ptr= NULL; //so we can detect if frame_end wasnt called (find some nicer solution...) | |
1510 } | |
1511 return buf_size; | |
1512 } | |
1513 | |
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
6481
diff
changeset
|
1514 av_cold int ff_rv34_decode_end(AVCodecContext *avctx) |
6026 | 1515 { |
1516 RV34DecContext *r = avctx->priv_data; | |
1517 | |
1518 MPV_common_end(&r->s); | |
1519 | |
1520 av_freep(&r->intra_types_hist); | |
1521 r->intra_types = NULL; | |
1522 av_freep(&r->mb_type); | |
8038 | 1523 av_freep(&r->cbp_luma); |
1524 av_freep(&r->cbp_chroma); | |
1525 av_freep(&r->deblock_coefs); | |
6026 | 1526 |
1527 return 0; | |
1528 } |