1 /*
2 * Copyright (C) 2004 The FFmpeg project
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file
23 * Standard C DSP-oriented functions cribbed from the original VP3
24 * source code.
25 */
26
27 #include "libavutil/attributes.h"
28 #include "libavutil/common.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/avassert.h"
31
32 #include "avcodec.h"
33 #include "rnd_avg.h"
34 #include "vp3dsp.h"
35
36 #define IdctAdjustBeforeShift 8
37 #define xC1S7 64277
38 #define xC2S6 60547
39 #define xC3S5 54491
40 #define xC4S4 46341
41 #define xC5S3 36410
42 #define xC6S2 25080
43 #define xC7S1 12785
44
45 #define M(a, b) ((int)((SUINT)(a) * (b)) >> 16)
46
idct(uint8_t * dst,ptrdiff_t stride,int16_t * input,int type)47 static av_always_inline void idct(uint8_t *dst, ptrdiff_t stride,
48 int16_t *input, int type)
49 {
50 int16_t *ip = input;
51
52 int A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;
53 int Ed, Gd, Add, Bdd, Fd, Hd;
54
55 int i;
56
57 /* Inverse DCT on the rows now */
58 for (i = 0; i < 8; i++) {
59 /* Check for non-zero values */
60 if (ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
61 ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8]) {
62 A = M(xC1S7, ip[1 * 8]) + M(xC7S1, ip[7 * 8]);
63 B = M(xC7S1, ip[1 * 8]) - M(xC1S7, ip[7 * 8]);
64 C = M(xC3S5, ip[3 * 8]) + M(xC5S3, ip[5 * 8]);
65 D = M(xC3S5, ip[5 * 8]) - M(xC5S3, ip[3 * 8]);
66
67 Ad = M(xC4S4, (A - C));
68 Bd = M(xC4S4, (B - D));
69
70 Cd = A + C;
71 Dd = B + D;
72
73 E = M(xC4S4, (ip[0 * 8] + ip[4 * 8]));
74 F = M(xC4S4, (ip[0 * 8] - ip[4 * 8]));
75
76 G = M(xC2S6, ip[2 * 8]) + M(xC6S2, ip[6 * 8]);
77 H = M(xC6S2, ip[2 * 8]) - M(xC2S6, ip[6 * 8]);
78
79 Ed = E - G;
80 Gd = E + G;
81
82 Add = F + Ad;
83 Bdd = Bd - H;
84
85 Fd = F - Ad;
86 Hd = Bd + H;
87
88 /* Final sequence of operations over-write original inputs. */
89 ip[0 * 8] = Gd + Cd;
90 ip[7 * 8] = Gd - Cd;
91
92 ip[1 * 8] = Add + Hd;
93 ip[2 * 8] = Add - Hd;
94
95 ip[3 * 8] = Ed + Dd;
96 ip[4 * 8] = Ed - Dd;
97
98 ip[5 * 8] = Fd + Bdd;
99 ip[6 * 8] = Fd - Bdd;
100 }
101
102 ip += 1; /* next row */
103 }
104
105 ip = input;
106
107 for (i = 0; i < 8; i++) {
108 /* Check for non-zero values (bitwise or faster than ||) */
109 if (ip[1] | ip[2] | ip[3] |
110 ip[4] | ip[5] | ip[6] | ip[7]) {
111 A = M(xC1S7, ip[1]) + M(xC7S1, ip[7]);
112 B = M(xC7S1, ip[1]) - M(xC1S7, ip[7]);
113 C = M(xC3S5, ip[3]) + M(xC5S3, ip[5]);
114 D = M(xC3S5, ip[5]) - M(xC5S3, ip[3]);
115
116 Ad = M(xC4S4, (A - C));
117 Bd = M(xC4S4, (B - D));
118
119 Cd = A + C;
120 Dd = B + D;
121
122 E = M(xC4S4, (ip[0] + ip[4])) + 8;
123 F = M(xC4S4, (ip[0] - ip[4])) + 8;
124
125 if (type == 1) { // HACK
126 E += 16 * 128;
127 F += 16 * 128;
128 }
129
130 G = M(xC2S6, ip[2]) + M(xC6S2, ip[6]);
131 H = M(xC6S2, ip[2]) - M(xC2S6, ip[6]);
132
133 Ed = E - G;
134 Gd = E + G;
135
136 Add = F + Ad;
137 Bdd = Bd - H;
138
139 Fd = F - Ad;
140 Hd = Bd + H;
141
142 /* Final sequence of operations over-write original inputs. */
143 if (type == 1) {
144 dst[0 * stride] = av_clip_uint8((Gd + Cd) >> 4);
145 dst[7 * stride] = av_clip_uint8((Gd - Cd) >> 4);
146
147 dst[1 * stride] = av_clip_uint8((Add + Hd) >> 4);
148 dst[2 * stride] = av_clip_uint8((Add - Hd) >> 4);
149
150 dst[3 * stride] = av_clip_uint8((Ed + Dd) >> 4);
151 dst[4 * stride] = av_clip_uint8((Ed - Dd) >> 4);
152
153 dst[5 * stride] = av_clip_uint8((Fd + Bdd) >> 4);
154 dst[6 * stride] = av_clip_uint8((Fd - Bdd) >> 4);
155 } else {
156 dst[0 * stride] = av_clip_uint8(dst[0 * stride] + ((Gd + Cd) >> 4));
157 dst[7 * stride] = av_clip_uint8(dst[7 * stride] + ((Gd - Cd) >> 4));
158
159 dst[1 * stride] = av_clip_uint8(dst[1 * stride] + ((Add + Hd) >> 4));
160 dst[2 * stride] = av_clip_uint8(dst[2 * stride] + ((Add - Hd) >> 4));
161
162 dst[3 * stride] = av_clip_uint8(dst[3 * stride] + ((Ed + Dd) >> 4));
163 dst[4 * stride] = av_clip_uint8(dst[4 * stride] + ((Ed - Dd) >> 4));
164
165 dst[5 * stride] = av_clip_uint8(dst[5 * stride] + ((Fd + Bdd) >> 4));
166 dst[6 * stride] = av_clip_uint8(dst[6 * stride] + ((Fd - Bdd) >> 4));
167 }
168 } else {
169 if (type == 1) {
170 dst[0*stride] =
171 dst[1*stride] =
172 dst[2*stride] =
173 dst[3*stride] =
174 dst[4*stride] =
175 dst[5*stride] =
176 dst[6*stride] =
177 dst[7*stride] = av_clip_uint8(128 + ((xC4S4 * ip[0] + (IdctAdjustBeforeShift << 16)) >> 20));
178 } else {
179 if (ip[0]) {
180 int v = (xC4S4 * ip[0] + (IdctAdjustBeforeShift << 16)) >> 20;
181 dst[0 * stride] = av_clip_uint8(dst[0 * stride] + v);
182 dst[1 * stride] = av_clip_uint8(dst[1 * stride] + v);
183 dst[2 * stride] = av_clip_uint8(dst[2 * stride] + v);
184 dst[3 * stride] = av_clip_uint8(dst[3 * stride] + v);
185 dst[4 * stride] = av_clip_uint8(dst[4 * stride] + v);
186 dst[5 * stride] = av_clip_uint8(dst[5 * stride] + v);
187 dst[6 * stride] = av_clip_uint8(dst[6 * stride] + v);
188 dst[7 * stride] = av_clip_uint8(dst[7 * stride] + v);
189 }
190 }
191 }
192
193 ip += 8; /* next column */
194 dst++;
195 }
196 }
197
idct10(uint8_t * dst,ptrdiff_t stride,int16_t * input,int type)198 static av_always_inline void idct10(uint8_t *dst, ptrdiff_t stride,
199 int16_t *input, int type)
200 {
201 int16_t *ip = input;
202
203 int A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;
204 int Ed, Gd, Add, Bdd, Fd, Hd;
205
206 int i;
207
208 /* Inverse DCT on the rows now */
209 for (i = 0; i < 4; i++) {
210 /* Check for non-zero values */
211 if (ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8]) {
212 A = M(xC1S7, ip[1 * 8]);
213 B = M(xC7S1, ip[1 * 8]);
214 C = M(xC3S5, ip[3 * 8]);
215 D = -M(xC5S3, ip[3 * 8]);
216
217 Ad = M(xC4S4, (A - C));
218 Bd = M(xC4S4, (B - D));
219
220 Cd = A + C;
221 Dd = B + D;
222
223 E = M(xC4S4, ip[0 * 8]);
224 F = E;
225
226 G = M(xC2S6, ip[2 * 8]);
227 H = M(xC6S2, ip[2 * 8]);
228
229 Ed = E - G;
230 Gd = E + G;
231
232 Add = F + Ad;
233 Bdd = Bd - H;
234
235 Fd = F - Ad;
236 Hd = Bd + H;
237
238 /* Final sequence of operations over-write original inputs */
239 ip[0 * 8] = Gd + Cd;
240 ip[7 * 8] = Gd - Cd;
241
242 ip[1 * 8] = Add + Hd;
243 ip[2 * 8] = Add - Hd;
244
245 ip[3 * 8] = Ed + Dd;
246 ip[4 * 8] = Ed - Dd;
247
248 ip[5 * 8] = Fd + Bdd;
249 ip[6 * 8] = Fd - Bdd;
250
251 }
252
253 ip += 1;
254 }
255
256 ip = input;
257
258 for (i = 0; i < 8; i++) {
259 /* Check for non-zero values (bitwise or faster than ||) */
260 if (ip[0] | ip[1] | ip[2] | ip[3]) {
261 A = M(xC1S7, ip[1]);
262 B = M(xC7S1, ip[1]);
263 C = M(xC3S5, ip[3]);
264 D = -M(xC5S3, ip[3]);
265
266 Ad = M(xC4S4, (A - C));
267 Bd = M(xC4S4, (B - D));
268
269 Cd = A + C;
270 Dd = B + D;
271
272 E = M(xC4S4, ip[0]);
273 if (type == 1)
274 E += 16 * 128;
275 F = E;
276
277 G = M(xC2S6, ip[2]);
278 H = M(xC6S2, ip[2]);
279
280 Ed = E - G;
281 Gd = E + G;
282
283 Add = F + Ad;
284 Bdd = Bd - H;
285
286 Fd = F - Ad;
287 Hd = Bd + H;
288
289 Gd += 8;
290 Add += 8;
291 Ed += 8;
292 Fd += 8;
293
294 /* Final sequence of operations over-write original inputs. */
295 if (type == 1) {
296 dst[0 * stride] = av_clip_uint8((Gd + Cd) >> 4);
297 dst[7 * stride] = av_clip_uint8((Gd - Cd) >> 4);
298
299 dst[1 * stride] = av_clip_uint8((Add + Hd) >> 4);
300 dst[2 * stride] = av_clip_uint8((Add - Hd) >> 4);
301
302 dst[3 * stride] = av_clip_uint8((Ed + Dd) >> 4);
303 dst[4 * stride] = av_clip_uint8((Ed - Dd) >> 4);
304
305 dst[5 * stride] = av_clip_uint8((Fd + Bdd) >> 4);
306 dst[6 * stride] = av_clip_uint8((Fd - Bdd) >> 4);
307 } else {
308 dst[0 * stride] = av_clip_uint8(dst[0 * stride] + ((Gd + Cd) >> 4));
309 dst[7 * stride] = av_clip_uint8(dst[7 * stride] + ((Gd - Cd) >> 4));
310
311 dst[1 * stride] = av_clip_uint8(dst[1 * stride] + ((Add + Hd) >> 4));
312 dst[2 * stride] = av_clip_uint8(dst[2 * stride] + ((Add - Hd) >> 4));
313
314 dst[3 * stride] = av_clip_uint8(dst[3 * stride] + ((Ed + Dd) >> 4));
315 dst[4 * stride] = av_clip_uint8(dst[4 * stride] + ((Ed - Dd) >> 4));
316
317 dst[5 * stride] = av_clip_uint8(dst[5 * stride] + ((Fd + Bdd) >> 4));
318 dst[6 * stride] = av_clip_uint8(dst[6 * stride] + ((Fd - Bdd) >> 4));
319 }
320 } else {
321 if (type == 1) {
322 dst[0*stride] =
323 dst[1*stride] =
324 dst[2*stride] =
325 dst[3*stride] =
326 dst[4*stride] =
327 dst[5*stride] =
328 dst[6*stride] =
329 dst[7*stride] = 128;
330 }
331 }
332
333 ip += 8;
334 dst++;
335 }
336 }
337
ff_vp3dsp_idct10_put(uint8_t * dest,ptrdiff_t stride,int16_t * block)338 void ff_vp3dsp_idct10_put(uint8_t *dest, ptrdiff_t stride, int16_t *block)
339 {
340 idct10(dest, stride, block, 1);
341 memset(block, 0, sizeof(*block) * 64);
342 }
343
ff_vp3dsp_idct10_add(uint8_t * dest,ptrdiff_t stride,int16_t * block)344 void ff_vp3dsp_idct10_add(uint8_t *dest, ptrdiff_t stride, int16_t *block)
345 {
346 idct10(dest, stride, block, 2);
347 memset(block, 0, sizeof(*block) * 64);
348 }
349
vp3_idct_put_c(uint8_t * dest,ptrdiff_t stride,int16_t * block)350 static void vp3_idct_put_c(uint8_t *dest /* align 8 */, ptrdiff_t stride,
351 int16_t *block /* align 16 */)
352 {
353 idct(dest, stride, block, 1);
354 memset(block, 0, sizeof(*block) * 64);
355 }
356
vp3_idct_add_c(uint8_t * dest,ptrdiff_t stride,int16_t * block)357 static void vp3_idct_add_c(uint8_t *dest /* align 8 */, ptrdiff_t stride,
358 int16_t *block /* align 16 */)
359 {
360 idct(dest, stride, block, 2);
361 memset(block, 0, sizeof(*block) * 64);
362 }
363
vp3_idct_dc_add_c(uint8_t * dest,ptrdiff_t stride,int16_t * block)364 static void vp3_idct_dc_add_c(uint8_t *dest /* align 8 */, ptrdiff_t stride,
365 int16_t *block /* align 16 */)
366 {
367 int i, dc = (block[0] + 15) >> 5;
368
369 for (i = 0; i < 8; i++) {
370 dest[0] = av_clip_uint8(dest[0] + dc);
371 dest[1] = av_clip_uint8(dest[1] + dc);
372 dest[2] = av_clip_uint8(dest[2] + dc);
373 dest[3] = av_clip_uint8(dest[3] + dc);
374 dest[4] = av_clip_uint8(dest[4] + dc);
375 dest[5] = av_clip_uint8(dest[5] + dc);
376 dest[6] = av_clip_uint8(dest[6] + dc);
377 dest[7] = av_clip_uint8(dest[7] + dc);
378 dest += stride;
379 }
380 block[0] = 0;
381 }
382
vp3_v_loop_filter_c(uint8_t * first_pixel,ptrdiff_t stride,int * bounding_values,int count)383 static av_always_inline void vp3_v_loop_filter_c(uint8_t *first_pixel, ptrdiff_t stride,
384 int *bounding_values, int count)
385 {
386 unsigned char *end;
387 int filter_value;
388 const ptrdiff_t nstride = -stride;
389
390 for (end = first_pixel + count; first_pixel < end; first_pixel++) {
391 filter_value = (first_pixel[2 * nstride] - first_pixel[stride]) +
392 (first_pixel[0] - first_pixel[nstride]) * 3;
393 filter_value = bounding_values[(filter_value + 4) >> 3];
394
395 first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value);
396 first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value);
397 }
398 }
399
vp3_h_loop_filter_c(uint8_t * first_pixel,ptrdiff_t stride,int * bounding_values,int count)400 static av_always_inline void vp3_h_loop_filter_c(uint8_t *first_pixel, ptrdiff_t stride,
401 int *bounding_values, int count)
402 {
403 unsigned char *end;
404 int filter_value;
405
406 for (end = first_pixel + count * stride; first_pixel != end; first_pixel += stride) {
407 filter_value = (first_pixel[-2] - first_pixel[1]) +
408 (first_pixel[ 0] - first_pixel[-1]) * 3;
409 filter_value = bounding_values[(filter_value + 4) >> 3];
410
411 first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
412 first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
413 }
414 }
415
416 #define LOOP_FILTER(prefix, suffix, dim, count) \
417 void prefix##_##dim##_loop_filter_##count##suffix(uint8_t *first_pixel, ptrdiff_t stride, \
418 int *bounding_values) \
419 { \
420 vp3_##dim##_loop_filter_c(first_pixel, stride, bounding_values, count); \
421 }
422
423 static LOOP_FILTER(vp3,_c, v, 8)
424 static LOOP_FILTER(vp3,_c, h, 8)
425 LOOP_FILTER(ff_vp3dsp, , v, 12)
426 LOOP_FILTER(ff_vp3dsp, , h, 12)
427
put_no_rnd_pixels_l2(uint8_t * dst,const uint8_t * src1,const uint8_t * src2,ptrdiff_t stride,int h)428 static void put_no_rnd_pixels_l2(uint8_t *dst, const uint8_t *src1,
429 const uint8_t *src2, ptrdiff_t stride, int h)
430 {
431 int i;
432
433 for (i = 0; i < h; i++) {
434 uint32_t a, b;
435
436 a = AV_RN32(&src1[i * stride]);
437 b = AV_RN32(&src2[i * stride]);
438 AV_WN32A(&dst[i * stride], no_rnd_avg32(a, b));
439 a = AV_RN32(&src1[i * stride + 4]);
440 b = AV_RN32(&src2[i * stride + 4]);
441 AV_WN32A(&dst[i * stride + 4], no_rnd_avg32(a, b));
442 }
443 }
444
ff_vp3dsp_init(VP3DSPContext * c,int flags)445 av_cold void ff_vp3dsp_init(VP3DSPContext *c, int flags)
446 {
447 c->put_no_rnd_pixels_l2 = put_no_rnd_pixels_l2;
448
449 c->idct_put = vp3_idct_put_c;
450 c->idct_add = vp3_idct_add_c;
451 c->idct_dc_add = vp3_idct_dc_add_c;
452 c->v_loop_filter = c->v_loop_filter_unaligned = vp3_v_loop_filter_8_c;
453 c->h_loop_filter = c->h_loop_filter_unaligned = vp3_h_loop_filter_8_c;
454
455 if (ARCH_ARM)
456 ff_vp3dsp_init_arm(c, flags);
457 if (ARCH_PPC)
458 ff_vp3dsp_init_ppc(c, flags);
459 if (ARCH_X86)
460 ff_vp3dsp_init_x86(c, flags);
461 if (ARCH_MIPS)
462 ff_vp3dsp_init_mips(c, flags);
463 }
464
465 /*
466 * This function initializes the loop filter boundary limits if the frame's
467 * quality index is different from the previous frame's.
468 *
469 * where sizeof(bounding_values_array) is 256 * sizeof(int)
470 *
471 * The filter_limit_values may not be larger than 127.
472 */
ff_vp3dsp_set_bounding_values(int * bounding_values_array,int filter_limit)473 void ff_vp3dsp_set_bounding_values(int * bounding_values_array, int filter_limit)
474 {
475 int *bounding_values = bounding_values_array + 127;
476 int x;
477 int value;
478
479 av_assert0(filter_limit < 128U);
480
481 /* set up the bounding values */
482 memset(bounding_values_array, 0, 256 * sizeof(int));
483 for (x = 0; x < filter_limit; x++) {
484 bounding_values[-x] = -x;
485 bounding_values[x] = x;
486 }
487 for (x = value = filter_limit; x < 128 && value; x++, value--) {
488 bounding_values[ x] = value;
489 bounding_values[-x] = -value;
490 }
491 if (value)
492 bounding_values[128] = value;
493 bounding_values[129] = bounding_values[130] = filter_limit * 0x02020202U;
494 }
495