2 * VC-1 and WMV3 decoder - DSP functions
3 * Copyright (c) 2006 Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
23 * @file libavcodec/vc1dsp.c
24 * VC-1 and WMV3 decoder
31 /** Apply overlap transform to horizontal edge
33 static void vc1_v_overlap_c(uint8_t* src, int stride)
39 for(i = 0; i < 8; i++) {
44 d1 = (a - d + 3 + rnd) >> 3;
45 d2 = (a - d + b - c + 4 - rnd) >> 3;
47 src[-2*stride] = a - d1;
48 src[-stride] = av_clip_uint8(b - d2);
49 src[0] = av_clip_uint8(c + d2);
56 /** Apply overlap transform to vertical edge
58 static void vc1_h_overlap_c(uint8_t* src, int stride)
64 for(i = 0; i < 8; i++) {
69 d1 = (a - d + 3 + rnd) >> 3;
70 d2 = (a - d + b - c + 4 - rnd) >> 3;
73 src[-1] = av_clip_uint8(b - d2);
74 src[0] = av_clip_uint8(c + d2);
82 * VC-1 in-loop deblocking filter for one line
83 * @param src source block type
84 * @param stride block stride
85 * @param pq block quantizer
86 * @return whether other 3 pairs should be filtered or not
89 static av_always_inline int vc1_filter_line(uint8_t* src, int stride, int pq){
90 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
92 int a0 = (2*(src[-2*stride] - src[ 1*stride]) - 5*(src[-1*stride] - src[ 0*stride]) + 4) >> 3;
93 int a0_sign = a0 >> 31; /* Store sign */
94 a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
96 int a1 = FFABS((2*(src[-4*stride] - src[-1*stride]) - 5*(src[-3*stride] - src[-2*stride]) + 4) >> 3);
97 int a2 = FFABS((2*(src[ 0*stride] - src[ 3*stride]) - 5*(src[ 1*stride] - src[ 2*stride]) + 4) >> 3);
98 if(a1 < a0 || a2 < a0){
99 int clip = src[-1*stride] - src[ 0*stride];
100 int clip_sign = clip >> 31;
101 clip = ((clip ^ clip_sign) - clip_sign)>>1;
103 int a3 = FFMIN(a1, a2);
104 int d = 5 * (a3 - a0);
105 int d_sign = (d >> 31);
106 d = ((d ^ d_sign) - d_sign) >> 3;
109 if( d_sign ^ clip_sign )
113 d = (d ^ d_sign) - d_sign; /* Restore sign */
114 src[-1*stride] = cm[src[-1*stride] - d];
115 src[ 0*stride] = cm[src[ 0*stride] + d];
125 * VC-1 in-loop deblocking filter
126 * @param src source block type
127 * @param step distance between horizontally adjacent elements
128 * @param stride distance between vertically adjacent elements
129 * @param len edge length to filter (4 or 8 pixels)
130 * @param pq block quantizer
133 static void vc1_loop_filter(uint8_t* src, int step, int stride, int len, int pq)
138 for(i = 0; i < len; i += 4){
139 filt3 = vc1_filter_line(src + 2*step, stride, pq);
141 vc1_filter_line(src + 0*step, stride, pq);
142 vc1_filter_line(src + 1*step, stride, pq);
143 vc1_filter_line(src + 3*step, stride, pq);
149 /** Do inverse transform on 8x8 block
151 static void vc1_inv_trans_8x8_c(DCTELEM block[64])
154 register int t1,t2,t3,t4,t5,t6,t7,t8;
159 for(i = 0; i < 8; i++){
160 t1 = 12 * (src[0] + src[4]) + 4;
161 t2 = 12 * (src[0] - src[4]) + 4;
162 t3 = 16 * src[2] + 6 * src[6];
163 t4 = 6 * src[2] - 16 * src[6];
170 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
171 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
172 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
173 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
175 dst[0] = (t5 + t1) >> 3;
176 dst[1] = (t6 + t2) >> 3;
177 dst[2] = (t7 + t3) >> 3;
178 dst[3] = (t8 + t4) >> 3;
179 dst[4] = (t8 - t4) >> 3;
180 dst[5] = (t7 - t3) >> 3;
181 dst[6] = (t6 - t2) >> 3;
182 dst[7] = (t5 - t1) >> 3;
190 for(i = 0; i < 8; i++){
191 t1 = 12 * (src[ 0] + src[32]) + 64;
192 t2 = 12 * (src[ 0] - src[32]) + 64;
193 t3 = 16 * src[16] + 6 * src[48];
194 t4 = 6 * src[16] - 16 * src[48];
201 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
202 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
203 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
204 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
206 dst[ 0] = (t5 + t1) >> 7;
207 dst[ 8] = (t6 + t2) >> 7;
208 dst[16] = (t7 + t3) >> 7;
209 dst[24] = (t8 + t4) >> 7;
210 dst[32] = (t8 - t4 + 1) >> 7;
211 dst[40] = (t7 - t3 + 1) >> 7;
212 dst[48] = (t6 - t2 + 1) >> 7;
213 dst[56] = (t5 - t1 + 1) >> 7;
220 /** Do inverse transform on 8x4 part of block
222 static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, DCTELEM *block)
225 register int t1,t2,t3,t4,t5,t6,t7,t8;
227 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
231 for(i = 0; i < 4; i++){
232 t1 = 12 * (src[0] + src[4]) + 4;
233 t2 = 12 * (src[0] - src[4]) + 4;
234 t3 = 16 * src[2] + 6 * src[6];
235 t4 = 6 * src[2] - 16 * src[6];
242 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
243 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
244 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
245 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
247 dst[0] = (t5 + t1) >> 3;
248 dst[1] = (t6 + t2) >> 3;
249 dst[2] = (t7 + t3) >> 3;
250 dst[3] = (t8 + t4) >> 3;
251 dst[4] = (t8 - t4) >> 3;
252 dst[5] = (t7 - t3) >> 3;
253 dst[6] = (t6 - t2) >> 3;
254 dst[7] = (t5 - t1) >> 3;
261 for(i = 0; i < 8; i++){
262 t1 = 17 * (src[ 0] + src[16]) + 64;
263 t2 = 17 * (src[ 0] - src[16]) + 64;
264 t3 = 22 * src[ 8] + 10 * src[24];
265 t4 = 22 * src[24] - 10 * src[ 8];
267 dest[0*linesize] = cm[dest[0*linesize] + ((t1 + t3) >> 7)];
268 dest[1*linesize] = cm[dest[1*linesize] + ((t2 - t4) >> 7)];
269 dest[2*linesize] = cm[dest[2*linesize] + ((t2 + t4) >> 7)];
270 dest[3*linesize] = cm[dest[3*linesize] + ((t1 - t3) >> 7)];
277 /** Do inverse transform on 4x8 parts of block
279 static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, DCTELEM *block)
282 register int t1,t2,t3,t4,t5,t6,t7,t8;
284 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
288 for(i = 0; i < 8; i++){
289 t1 = 17 * (src[0] + src[2]) + 4;
290 t2 = 17 * (src[0] - src[2]) + 4;
291 t3 = 22 * src[1] + 10 * src[3];
292 t4 = 22 * src[3] - 10 * src[1];
294 dst[0] = (t1 + t3) >> 3;
295 dst[1] = (t2 - t4) >> 3;
296 dst[2] = (t2 + t4) >> 3;
297 dst[3] = (t1 - t3) >> 3;
304 for(i = 0; i < 4; i++){
305 t1 = 12 * (src[ 0] + src[32]) + 64;
306 t2 = 12 * (src[ 0] - src[32]) + 64;
307 t3 = 16 * src[16] + 6 * src[48];
308 t4 = 6 * src[16] - 16 * src[48];
315 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
316 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
317 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
318 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
320 dest[0*linesize] = cm[dest[0*linesize] + ((t5 + t1) >> 7)];
321 dest[1*linesize] = cm[dest[1*linesize] + ((t6 + t2) >> 7)];
322 dest[2*linesize] = cm[dest[2*linesize] + ((t7 + t3) >> 7)];
323 dest[3*linesize] = cm[dest[3*linesize] + ((t8 + t4) >> 7)];
324 dest[4*linesize] = cm[dest[4*linesize] + ((t8 - t4 + 1) >> 7)];
325 dest[5*linesize] = cm[dest[5*linesize] + ((t7 - t3 + 1) >> 7)];
326 dest[6*linesize] = cm[dest[6*linesize] + ((t6 - t2 + 1) >> 7)];
327 dest[7*linesize] = cm[dest[7*linesize] + ((t5 - t1 + 1) >> 7)];
334 /** Do inverse transform on 4x4 part of block
336 static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, DCTELEM *block)
339 register int t1,t2,t3,t4;
341 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
345 for(i = 0; i < 4; i++){
346 t1 = 17 * (src[0] + src[2]) + 4;
347 t2 = 17 * (src[0] - src[2]) + 4;
348 t3 = 22 * src[1] + 10 * src[3];
349 t4 = 22 * src[3] - 10 * src[1];
351 dst[0] = (t1 + t3) >> 3;
352 dst[1] = (t2 - t4) >> 3;
353 dst[2] = (t2 + t4) >> 3;
354 dst[3] = (t1 - t3) >> 3;
361 for(i = 0; i < 4; i++){
362 t1 = 17 * (src[ 0] + src[16]) + 64;
363 t2 = 17 * (src[ 0] - src[16]) + 64;
364 t3 = 22 * src[ 8] + 10 * src[24];
365 t4 = 22 * src[24] - 10 * src[ 8];
367 dest[0*linesize] = cm[dest[0*linesize] + ((t1 + t3) >> 7)];
368 dest[1*linesize] = cm[dest[1*linesize] + ((t2 - t4) >> 7)];
369 dest[2*linesize] = cm[dest[2*linesize] + ((t2 + t4) >> 7)];
370 dest[3*linesize] = cm[dest[3*linesize] + ((t1 - t3) >> 7)];
377 /* motion compensation functions */
378 /** Filter in case of 2 filters */
379 #define VC1_MSPEL_FILTER_16B(DIR, TYPE) \
380 static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, int stride, int mode) \
383 case 0: /* no shift - should not occur */ \
385 case 1: /* 1/4 shift */ \
386 return -4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2]; \
387 case 2: /* 1/2 shift */ \
388 return -src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2]; \
389 case 3: /* 3/4 shift */ \
390 return -3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2]; \
392 return 0; /* should not occur */ \
395 VC1_MSPEL_FILTER_16B(ver, uint8_t);
396 VC1_MSPEL_FILTER_16B(hor, int16_t);
399 /** Filter used to interpolate fractional pel values
401 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride, int mode, int r)
407 return (-4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2] + 32 - r) >> 6;
409 return (-src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2] + 8 - r) >> 4;
411 return (-3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2] + 32 - r) >> 6;
413 return 0; //should not occur
416 /** Function used to do motion compensation with bicubic interpolation
418 #define VC1_MSPEL_MC(OP, OPNAME)\
419 static void OPNAME ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride, int hmode, int vmode, int rnd)\
423 if (vmode) { /* Horizontal filter to apply */\
426 if (hmode) { /* Vertical filter to apply, output to tmp */\
427 static const int shift_value[] = { 0, 5, 1, 5 };\
428 int shift = (shift_value[hmode]+shift_value[vmode])>>1;\
429 int16_t tmp[11*8], *tptr = tmp;\
431 r = (1<<(shift-1)) + rnd-1;\
434 for(j = 0; j < 8; j++) {\
435 for(i = 0; i < 11; i++)\
436 tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode)+r)>>shift;\
443 for(j = 0; j < 8; j++) {\
444 for(i = 0; i < 8; i++)\
445 OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode)+r)>>7);\
452 else { /* No horizontal filter, output 8 lines to dst */\
455 for(j = 0; j < 8; j++) {\
456 for(i = 0; i < 8; i++)\
457 OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r));\
465 /* Horizontal mode with no vertical mode */\
466 for(j = 0; j < 8; j++) {\
467 for(i = 0; i < 8; i++)\
468 OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd));\
474 #define op_put(a, b) a = av_clip_uint8(b)
475 #define op_avg(a, b) a = (a + av_clip_uint8(b) + 1) >> 1
477 VC1_MSPEL_MC(op_put, put_)
478 VC1_MSPEL_MC(op_avg, avg_)
480 /* pixel functions - really are entry points to vc1_mspel_mc */
482 /* this one is defined in dsputil.c */
483 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
484 void ff_avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
486 #define PUT_VC1_MSPEL(a, b)\
487 static void put_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
488 put_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
490 static void avg_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
491 avg_vc1_mspel_mc(dst, src, stride, a, b, rnd); \
513 void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
514 dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
515 dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
516 dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;
517 dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;
518 dsp->vc1_h_overlap = vc1_h_overlap_c;
519 dsp->vc1_v_overlap = vc1_v_overlap_c;
520 dsp->vc1_loop_filter = vc1_loop_filter;
522 dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_c;
523 dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_c;
524 dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_c;
525 dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_c;
526 dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_c;
527 dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_c;
528 dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_c;
529 dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_c;
530 dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_c;
531 dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_c;
532 dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_c;
533 dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_c;
534 dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_c;
535 dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_c;
536 dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_c;
537 dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_c;
539 dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_vc1_mspel_mc00_c;
540 dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_c;
541 dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_c;
542 dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_c;
543 dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_c;
544 dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_c;
545 dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_c;
546 dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_c;
547 dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_c;
548 dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_c;
549 dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_c;
550 dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_c;
551 dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_c;
552 dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_c;
553 dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_c;
554 dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_c;