2
2
* VC-1 and WMV3 decoder - DSP functions
3
3
* Copyright (c) 2006 Konstantin Shishkov
5
* This file is part of FFmpeg.
5
* This file is part of Libav.
7
* FFmpeg is free software; you can redistribute it and/or
7
* Libav is free software; you can redistribute it and/or
8
8
* modify it under the terms of the GNU Lesser General Public
9
9
* License as published by the Free Software Foundation; either
10
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,
12
* Libav is distributed in the hope that it will be useful,
13
13
* but WITHOUT ANY WARRANTY; without even the implied warranty of
14
14
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15
15
* Lesser General Public License for more details.
17
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
18
* License along with Libav; if not, write to the Free Software
19
19
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
184
184
int dc = block[0];
185
const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
186
186
dc = (3 * dc + 1) >> 1;
187
187
dc = (3 * dc + 16) >> 5;
188
cm = ff_cropTbl + MAX_NEG_CROP + dc;
188
189
for(i = 0; i < 8; i++){
189
dest[0] = cm[dest[0]+dc];
190
dest[1] = cm[dest[1]+dc];
191
dest[2] = cm[dest[2]+dc];
192
dest[3] = cm[dest[3]+dc];
193
dest[4] = cm[dest[4]+dc];
194
dest[5] = cm[dest[5]+dc];
195
dest[6] = cm[dest[6]+dc];
196
dest[7] = cm[dest[7]+dc];
190
dest[0] = cm[dest[0]];
191
dest[1] = cm[dest[1]];
192
dest[2] = cm[dest[2]];
193
dest[3] = cm[dest[3]];
194
dest[4] = cm[dest[4]];
195
dest[5] = cm[dest[5]];
196
dest[6] = cm[dest[6]];
197
dest[7] = cm[dest[7]];
197
198
dest += linesize;
201
static void vc1_inv_trans_8x8_c(DCTELEM block[64])
202
static av_always_inline void vc1_inv_trans_8x8_c(DCTELEM block[64], int shl, int sub)
204
205
register int t1,t2,t3,t4,t5,t6,t7,t8;
206
DCTELEM *src, *dst, temp[64];
209
210
for(i = 0; i < 8; i++){
210
t1 = 12 * (src[0] + src[4]) + 4;
211
t2 = 12 * (src[0] - src[4]) + 4;
212
t3 = 16 * src[2] + 6 * src[6];
213
t4 = 6 * src[2] - 16 * src[6];
211
t1 = 12 * (src[ 0] + src[32]) + 4;
212
t2 = 12 * (src[ 0] - src[32]) + 4;
213
t3 = 16 * src[16] + 6 * src[48];
214
t4 = 6 * src[16] - 16 * src[48];
220
t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7];
221
t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7];
222
t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7];
223
t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7];
221
t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56];
222
t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56];
223
t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
224
t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
225
226
dst[0] = (t5 + t1) >> 3;
226
227
dst[1] = (t6 + t2) >> 3;
253
254
t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56];
254
255
t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56];
256
dst[ 0] = (t5 + t1) >> 7;
257
dst[ 8] = (t6 + t2) >> 7;
258
dst[16] = (t7 + t3) >> 7;
259
dst[24] = (t8 + t4) >> 7;
260
dst[32] = (t8 - t4 + 1) >> 7;
261
dst[40] = (t7 - t3 + 1) >> 7;
262
dst[48] = (t6 - t2 + 1) >> 7;
263
dst[56] = (t5 - t1 + 1) >> 7;
257
dst[ 0] = (((t5 + t1 ) >> 7) - sub) << shl;
258
dst[ 8] = (((t6 + t2 ) >> 7) - sub) << shl;
259
dst[16] = (((t7 + t3 ) >> 7) - sub) << shl;
260
dst[24] = (((t8 + t4 ) >> 7) - sub) << shl;
261
dst[32] = (((t8 - t4 + 1) >> 7) - sub) << shl;
262
dst[40] = (((t7 - t3 + 1) >> 7) - sub) << shl;
263
dst[48] = (((t6 - t2 + 1) >> 7) - sub) << shl;
264
dst[56] = (((t5 - t1 + 1) >> 7) - sub) << shl;
271
static void vc1_inv_trans_8x8_add_c(uint8_t *dest, int linesize, DCTELEM *block)
273
vc1_inv_trans_8x8_c(block, 0, 0);
274
ff_add_pixels_clamped_c(block, dest, linesize);
277
static void vc1_inv_trans_8x8_put_signed_c(uint8_t *dest, int linesize, DCTELEM *block)
279
vc1_inv_trans_8x8_c(block, 0, 0);
280
ff_put_signed_pixels_clamped_c(block, dest, linesize);
283
static void vc1_inv_trans_8x8_put_signed_rangered_c(uint8_t *dest, int linesize, DCTELEM *block)
285
vc1_inv_trans_8x8_c(block, 1, 0);
286
ff_put_signed_pixels_clamped_c(block, dest, linesize);
289
static void vc1_inv_trans_8x8_put_c(uint8_t *dest, int linesize, DCTELEM *block)
291
vc1_inv_trans_8x8_c(block, 0, 0);
292
ff_put_pixels_clamped_c(block, dest, linesize);
295
static void vc1_inv_trans_8x8_put_rangered_c(uint8_t *dest, int linesize, DCTELEM *block)
297
vc1_inv_trans_8x8_c(block, 1, 64);
298
ff_put_pixels_clamped_c(block, dest, linesize);
270
301
/** Do inverse transform on 8x4 part of block
272
303
static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, DCTELEM *block)
275
306
int dc = block[0];
276
const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
277
308
dc = ( 3 * dc + 1) >> 1;
278
309
dc = (17 * dc + 64) >> 7;
310
cm = ff_cropTbl + MAX_NEG_CROP + dc;
279
311
for(i = 0; i < 4; i++){
280
dest[0] = cm[dest[0]+dc];
281
dest[1] = cm[dest[1]+dc];
282
dest[2] = cm[dest[2]+dc];
283
dest[3] = cm[dest[3]+dc];
284
dest[4] = cm[dest[4]+dc];
285
dest[5] = cm[dest[5]+dc];
286
dest[6] = cm[dest[6]+dc];
287
dest[7] = cm[dest[7]+dc];
312
dest[0] = cm[dest[0]];
313
dest[1] = cm[dest[1]];
314
dest[2] = cm[dest[2]];
315
dest[3] = cm[dest[3]];
316
dest[4] = cm[dest[4]];
317
dest[5] = cm[dest[5]];
318
dest[6] = cm[dest[6]];
319
dest[7] = cm[dest[7]];
288
320
dest += linesize;
608
642
PUT_VC1_MSPEL(2, 3)
609
643
PUT_VC1_MSPEL(3, 3)
611
av_cold void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) {
612
dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
645
static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
646
const int A=(8-x)*(8-y);
647
const int B=( x)*(8-y);
648
const int C=(8-x)*( y);
649
const int D=( x)*( y);
652
assert(x<8 && y<8 && x>=0 && y>=0);
656
dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6;
657
dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6;
658
dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6;
659
dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6;
660
dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6;
661
dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6;
662
dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6;
663
dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6;
669
#define avg2(a,b) ((a+b+1)>>1)
670
static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
671
const int A=(8-x)*(8-y);
672
const int B=( x)*(8-y);
673
const int C=(8-x)*( y);
674
const int D=( x)*( y);
677
assert(x<8 && y<8 && x>=0 && y>=0);
681
dst[0] = avg2(dst[0], ((A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6));
682
dst[1] = avg2(dst[1], ((A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6));
683
dst[2] = avg2(dst[2], ((A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6));
684
dst[3] = avg2(dst[3], ((A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6));
685
dst[4] = avg2(dst[4], ((A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6));
686
dst[5] = avg2(dst[5], ((A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6));
687
dst[6] = avg2(dst[6], ((A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6));
688
dst[7] = avg2(dst[7], ((A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6));
694
av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {
695
dsp->vc1_inv_trans_8x8_add = vc1_inv_trans_8x8_add_c;
696
dsp->vc1_inv_trans_8x8_put_signed[0] = vc1_inv_trans_8x8_put_signed_c;
697
dsp->vc1_inv_trans_8x8_put_signed[1] = vc1_inv_trans_8x8_put_signed_rangered_c;
698
dsp->vc1_inv_trans_8x8_put[0] = vc1_inv_trans_8x8_put_c;
699
dsp->vc1_inv_trans_8x8_put[1] = vc1_inv_trans_8x8_put_rangered_c;
613
700
dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
614
701
dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;
615
702
dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;