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
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] = b - 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;
82 /** Do inverse transform on 8x8 block
84 static void vc1_inv_trans_8x8_c(DCTELEM block[64])
87 register int t1,t2,t3,t4,t5,t6,t7,t8;
92 for(i = 0; i < 8; i++){
93 t1 = 12 * (src[0] + src[4]);
94 t2 = 12 * (src[0] - src[4]);
95 t3 = 16 * src[2] + 6 * src[6];
96 t4 = 6 * src[2] - 16 * src[6];
103 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
104 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
105 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
106 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
108 dst[0] = (t5 + t1 + 4) >> 3;
109 dst[1] = (t6 + t2 + 4) >> 3;
110 dst[2] = (t7 + t3 + 4) >> 3;
111 dst[3] = (t8 + t4 + 4) >> 3;
112 dst[4] = (t8 - t4 + 4) >> 3;
113 dst[5] = (t7 - t3 + 4) >> 3;
114 dst[6] = (t6 - t2 + 4) >> 3;
115 dst[7] = (t5 - t1 + 4) >> 3;
123 for(i = 0; i < 8; i++){
124 t1 = 12 * (src[ 0] + src[32]);
125 t2 = 12 * (src[ 0] - src[32]);
126 t3 = 16 * src[16] + 6 * src[48];
127 t4 = 6 * src[16] - 16 * src[48];
134 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
135 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
136 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
137 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
139 dst[ 0] = (t5 + t1 + 64) >> 7;
140 dst[ 8] = (t6 + t2 + 64) >> 7;
141 dst[16] = (t7 + t3 + 64) >> 7;
142 dst[24] = (t8 + t4 + 64) >> 7;
143 dst[32] = (t8 - t4 + 64 + 1) >> 7;
144 dst[40] = (t7 - t3 + 64 + 1) >> 7;
145 dst[48] = (t6 - t2 + 64 + 1) >> 7;
146 dst[56] = (t5 - t1 + 64 + 1) >> 7;
153 /** Do inverse transform on 8x4 part of block
155 static void vc1_inv_trans_8x4_c(DCTELEM block[64], int n)
158 register int t1,t2,t3,t4,t5,t6,t7,t8;
165 for(i = 0; i < 4; i++){
166 t1 = 12 * (src[0] + src[4]);
167 t2 = 12 * (src[0] - src[4]);
168 t3 = 16 * src[2] + 6 * src[6];
169 t4 = 6 * src[2] - 16 * src[6];
176 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
177 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
178 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
179 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
181 dst[0] = (t5 + t1 + 4) >> 3;
182 dst[1] = (t6 + t2 + 4) >> 3;
183 dst[2] = (t7 + t3 + 4) >> 3;
184 dst[3] = (t8 + t4 + 4) >> 3;
185 dst[4] = (t8 - t4 + 4) >> 3;
186 dst[5] = (t7 - t3 + 4) >> 3;
187 dst[6] = (t6 - t2 + 4) >> 3;
188 dst[7] = (t5 - t1 + 4) >> 3;
196 for(i = 0; i < 8; i++){
197 t1 = 17 * (src[ 0] + src[16]);
198 t2 = 17 * (src[ 0] - src[16]);
204 dst[ 0] = (t1 + t3 + t6 + 64) >> 7;
205 dst[ 8] = (t2 - t4 + t5 + 64) >> 7;
206 dst[16] = (t2 + t4 - t5 + 64) >> 7;
207 dst[24] = (t1 - t3 - t6 + 64) >> 7;
214 /** Do inverse transform on 4x8 parts of block
216 static void vc1_inv_trans_4x8_c(DCTELEM block[64], int n)
219 register int t1,t2,t3,t4,t5,t6,t7,t8;
226 for(i = 0; i < 8; i++){
227 t1 = 17 * (src[0] + src[2]);
228 t2 = 17 * (src[0] - src[2]);
234 dst[0] = (t1 + t3 + t6 + 4) >> 3;
235 dst[1] = (t2 - t4 + t5 + 4) >> 3;
236 dst[2] = (t2 + t4 - t5 + 4) >> 3;
237 dst[3] = (t1 - t3 - t6 + 4) >> 3;
245 for(i = 0; i < 4; i++){
246 t1 = 12 * (src[ 0] + src[32]);
247 t2 = 12 * (src[ 0] - src[32]);
248 t3 = 16 * src[16] + 6 * src[48];
249 t4 = 6 * src[16] - 16 * src[48];
256 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
257 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
258 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
259 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
261 dst[ 0] = (t5 + t1 + 64) >> 7;
262 dst[ 8] = (t6 + t2 + 64) >> 7;
263 dst[16] = (t7 + t3 + 64) >> 7;
264 dst[24] = (t8 + t4 + 64) >> 7;
265 dst[32] = (t8 - t4 + 64 + 1) >> 7;
266 dst[40] = (t7 - t3 + 64 + 1) >> 7;
267 dst[48] = (t6 - t2 + 64 + 1) >> 7;
268 dst[56] = (t5 - t1 + 64 + 1) >> 7;
275 /** Do inverse transform on 4x4 part of block
277 static void vc1_inv_trans_4x4_c(DCTELEM block[64], int n)
280 register int t1,t2,t3,t4,t5,t6;
284 off = (n&1) * 4 + (n&2) * 16;
287 for(i = 0; i < 4; i++){
288 t1 = 17 * (src[0] + src[2]);
289 t2 = 17 * (src[0] - src[2]);
295 dst[0] = (t1 + t3 + t6 + 4) >> 3;
296 dst[1] = (t2 - t4 + t5 + 4) >> 3;
297 dst[2] = (t2 + t4 - t5 + 4) >> 3;
298 dst[3] = (t1 - t3 - t6 + 4) >> 3;
306 for(i = 0; i < 4; i++){
307 t1 = 17 * (src[ 0] + src[16]);
308 t2 = 17 * (src[ 0] - src[16]);
314 dst[ 0] = (t1 + t3 + t6 + 64) >> 7;
315 dst[ 8] = (t2 - t4 + t5 + 64) >> 7;
316 dst[16] = (t2 + t4 - t5 + 64) >> 7;
317 dst[24] = (t1 - t3 - t6 + 64) >> 7;
324 /* motion compensation functions */
325 /** Filter in case of 2 filters */
326 #define VC1_MSPEL_FILTER_16B(DIR, TYPE) \
327 static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, int stride, int mode) \
330 case 0: /* no shift - should not occur */ \
332 case 1: /* 1/4 shift */ \
333 return -4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2]; \
334 case 2: /* 1/2 shift */ \
335 return -src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2]; \
336 case 3: /* 3/4 shift */ \
337 return -3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2]; \
339 return 0; /* should not occur */ \
342 VC1_MSPEL_FILTER_16B(ver, uint8_t);
343 VC1_MSPEL_FILTER_16B(hor, int16_t);
346 /** Filter used to interpolate fractional pel values
348 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride, int mode, int r)
354 return (-4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2] + 32 - r) >> 6;
356 return (-src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2] + 8 - r) >> 4;
358 return (-3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2] + 32 - r) >> 6;
360 return 0; //should not occur
363 /** Function used to do motion compensation with bicubic interpolation
365 static void vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride, int hmode, int vmode, int rnd)
369 if (vmode) { /* Horizontal filter to apply */
372 if (hmode) { /* Vertical filter to apply, output to tmp */
373 static const int shift_value[] = { 0, 5, 1, 5 };
374 int shift = (shift_value[hmode]+shift_value[vmode])>>1;
375 int16_t tmp[11*8], *tptr = tmp;
377 r = (1<<(shift-1)) + rnd-1;
380 for(j = 0; j < 8; j++) {
381 for(i = 0; i < 11; i++)
382 tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode)+r)>>shift;
389 for(j = 0; j < 8; j++) {
390 for(i = 0; i < 8; i++)
391 dst[i] = av_clip_uint8((vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode)+r)>>7);
398 else { /* No horizontal filter, output 8 lines to dst */
401 for(j = 0; j < 8; j++) {
402 for(i = 0; i < 8; i++)
403 dst[i] = av_clip_uint8(vc1_mspel_filter(src + i, stride, vmode, r));
411 /* Horizontal mode with no vertical mode */
412 for(j = 0; j < 8; j++) {
413 for(i = 0; i < 8; i++)
414 dst[i] = av_clip_uint8(vc1_mspel_filter(src + i, 1, hmode, rnd));
420 /* pixel functions - really are entry points to vc1_mspel_mc */
422 /* this one is defined in dsputil.c */
423 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd);
425 #define PUT_VC1_MSPEL(a, b)\
426 static void put_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
427 vc1_mspel_mc(dst, src, stride, a, b, rnd); \
449 void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
450 dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
451 dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
452 dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;
453 dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;
454 dsp->vc1_h_overlap = vc1_h_overlap_c;
455 dsp->vc1_v_overlap = vc1_v_overlap_c;
457 dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_c;
458 dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_c;
459 dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_c;
460 dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_c;
461 dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_c;
462 dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_c;
463 dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_c;
464 dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_c;
465 dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_c;
466 dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_c;
467 dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_c;
468 dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_c;
469 dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_c;
470 dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_c;
471 dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_c;
472 dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_c;