1 /*
2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Clément Bœsch <u pkh me>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (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
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20 */
21
22 /**
23 * @file
24 * Simple post processing filter
25 *
26 * This implementation is based on an algorithm described in
27 * "Aria Nosratinia Embedded Post-Processing for
28 * Enhancement of Compressed Images (1999)"
29 *
30 * Originally written by Michael Niedermayer for the MPlayer project, and
31 * ported by Clément Bœsch for FFmpeg.
32 */
33
34 #include "libavutil/avassert.h"
35 #include "libavutil/imgutils.h"
36 #include "libavutil/mem_internal.h"
37 #include "libavutil/opt.h"
38 #include "libavutil/pixdesc.h"
39 #include "internal.h"
40 #include "qp_table.h"
41 #include "vf_spp.h"
42
43 enum mode {
44 MODE_HARD,
45 MODE_SOFT,
46 NB_MODES
47 };
48
49 #if FF_API_CHILD_CLASS_NEXT
child_class_next(const AVClass * prev)50 static const AVClass *child_class_next(const AVClass *prev)
51 {
52 return prev ? NULL : avcodec_dct_get_class();
53 }
54 #endif
55
child_class_iterate(void ** iter)56 static const AVClass *child_class_iterate(void **iter)
57 {
58 const AVClass *c = *iter ? NULL : avcodec_dct_get_class();
59 *iter = (void*)(uintptr_t)c;
60 return c;
61 }
62
child_next(void * obj,void * prev)63 static void *child_next(void *obj, void *prev)
64 {
65 SPPContext *s = obj;
66 return prev ? NULL : s->dct;
67 }
68
69 #define OFFSET(x) offsetof(SPPContext, x)
70 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
71 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
72 static const AVOption spp_options[] = {
73 { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, TFLAGS },
74 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
75 { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
76 { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
77 { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
78 { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS },
79 { NULL }
80 };
81
82 static const AVClass spp_class = {
83 .class_name = "spp",
84 .item_name = av_default_item_name,
85 .option = spp_options,
86 .version = LIBAVUTIL_VERSION_INT,
87 .category = AV_CLASS_CATEGORY_FILTER,
88 #if FF_API_CHILD_CLASS_NEXT
89 .child_class_next = child_class_next,
90 #endif
91 .child_class_iterate = child_class_iterate,
92 .child_next = child_next,
93 };
94
95 // XXX: share between filters?
96 DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
97 { 0, 48, 12, 60, 3, 51, 15, 63 },
98 { 32, 16, 44, 28, 35, 19, 47, 31 },
99 { 8, 56, 4, 52, 11, 59, 7, 55 },
100 { 40, 24, 36, 20, 43, 27, 39, 23 },
101 { 2, 50, 14, 62, 1, 49, 13, 61 },
102 { 34, 18, 46, 30, 33, 17, 45, 29 },
103 { 10, 58, 6, 54, 9, 57, 5, 53 },
104 { 42, 26, 38, 22, 41, 25, 37, 21 },
105 };
106
107 static const uint8_t offset[127][2] = {
108 {0,0},
109 {0,0}, {4,4}, // quality = 1
110 {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2
111 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
112
113 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
114 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
115
116 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
117 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
118 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
119 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
120
121 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
122 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
123 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
124 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
125 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
126 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
127 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
128 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
129 };
130
hardthresh_c(int16_t dst[64],const int16_t src[64],int qp,const uint8_t * permutation)131 static void hardthresh_c(int16_t dst[64], const int16_t src[64],
132 int qp, const uint8_t *permutation)
133 {
134 int i;
135 int bias = 0; // FIXME
136
137 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
138 unsigned threshold2 = threshold1 << 1;
139
140 memset(dst, 0, 64 * sizeof(dst[0]));
141 dst[0] = (src[0] + 4) >> 3;
142
143 for (i = 1; i < 64; i++) {
144 int level = src[i];
145 if (((unsigned)(level + threshold1)) > threshold2) {
146 const int j = permutation[i];
147 dst[j] = (level + 4) >> 3;
148 }
149 }
150 }
151
softthresh_c(int16_t dst[64],const int16_t src[64],int qp,const uint8_t * permutation)152 static void softthresh_c(int16_t dst[64], const int16_t src[64],
153 int qp, const uint8_t *permutation)
154 {
155 int i;
156 int bias = 0; //FIXME
157
158 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
159 unsigned threshold2 = threshold1 << 1;
160
161 memset(dst, 0, 64 * sizeof(dst[0]));
162 dst[0] = (src[0] + 4) >> 3;
163
164 for (i = 1; i < 64; i++) {
165 int level = src[i];
166 if (((unsigned)(level + threshold1)) > threshold2) {
167 const int j = permutation[i];
168 if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
169 else dst[j] = (level + threshold1 + 4) >> 3;
170 }
171 }
172 }
173
store_slice_c(uint8_t * dst,const int16_t * src,int dst_linesize,int src_linesize,int width,int height,int log2_scale,const uint8_t dither[8][8])174 static void store_slice_c(uint8_t *dst, const int16_t *src,
175 int dst_linesize, int src_linesize,
176 int width, int height, int log2_scale,
177 const uint8_t dither[8][8])
178 {
179 int y, x;
180
181 #define STORE(pos) do { \
182 temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \
183 if (temp & 0x100) \
184 temp = ~(temp >> 31); \
185 dst[x + y*dst_linesize + pos] = temp; \
186 } while (0)
187
188 for (y = 0; y < height; y++) {
189 const uint8_t *d = dither[y];
190 for (x = 0; x < width; x += 8) {
191 int temp;
192 STORE(0);
193 STORE(1);
194 STORE(2);
195 STORE(3);
196 STORE(4);
197 STORE(5);
198 STORE(6);
199 STORE(7);
200 }
201 }
202 }
203
store_slice16_c(uint16_t * dst,const int16_t * src,int dst_linesize,int src_linesize,int width,int height,int log2_scale,const uint8_t dither[8][8],int depth)204 static void store_slice16_c(uint16_t *dst, const int16_t *src,
205 int dst_linesize, int src_linesize,
206 int width, int height, int log2_scale,
207 const uint8_t dither[8][8], int depth)
208 {
209 int y, x;
210 unsigned int mask = -1<<depth;
211
212 #define STORE16(pos) do { \
213 temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5; \
214 if (temp & mask ) \
215 temp = ~(temp >> 31); \
216 dst[x + y*dst_linesize + pos] = temp; \
217 } while (0)
218
219 for (y = 0; y < height; y++) {
220 const uint8_t *d = dither[y];
221 for (x = 0; x < width; x += 8) {
222 int temp;
223 STORE16(0);
224 STORE16(1);
225 STORE16(2);
226 STORE16(3);
227 STORE16(4);
228 STORE16(5);
229 STORE16(6);
230 STORE16(7);
231 }
232 }
233 }
234
add_block(uint16_t * dst,int linesize,const int16_t block[64])235 static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
236 {
237 int y;
238
239 for (y = 0; y < 8; y++) {
240 dst[0 + y*linesize] += block[0 + y*8];
241 dst[1 + y*linesize] += block[1 + y*8];
242 dst[2 + y*linesize] += block[2 + y*8];
243 dst[3 + y*linesize] += block[3 + y*8];
244 dst[4 + y*linesize] += block[4 + y*8];
245 dst[5 + y*linesize] += block[5 + y*8];
246 dst[6 + y*linesize] += block[6 + y*8];
247 dst[7 + y*linesize] += block[7 + y*8];
248 }
249 }
250
filter(SPPContext * p,uint8_t * dst,uint8_t * src,int dst_linesize,int src_linesize,int width,int height,const uint8_t * qp_table,int qp_stride,int is_luma,int depth)251 static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
252 int dst_linesize, int src_linesize, int width, int height,
253 const uint8_t *qp_table, int qp_stride, int is_luma, int depth)
254 {
255 int x, y, i;
256 const int count = 1 << p->log2_count;
257 const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
258 DECLARE_ALIGNED(16, uint64_t, block_align)[32];
259 int16_t *block = (int16_t *)block_align;
260 int16_t *block2 = (int16_t *)(block_align + 16);
261 uint16_t *psrc16 = (uint16_t*)p->src;
262 const int sample_bytes = (depth+7) / 8;
263
264 for (y = 0; y < height; y++) {
265 int index = 8 + 8*linesize + y*linesize;
266 memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
267 if (sample_bytes == 1) {
268 for (x = 0; x < 8; x++) {
269 p->src[index - x - 1] = p->src[index + x ];
270 p->src[index + width + x ] = p->src[index + width - x - 1];
271 }
272 } else {
273 for (x = 0; x < 8; x++) {
274 psrc16[index - x - 1] = psrc16[index + x ];
275 psrc16[index + width + x ] = psrc16[index + width - x - 1];
276 }
277 }
278 }
279 for (y = 0; y < 8; y++) {
280 memcpy(p->src + ( 7-y)*linesize * sample_bytes, p->src + ( y+8)*linesize * sample_bytes, linesize * sample_bytes);
281 memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
282 }
283
284 for (y = 0; y < height + 8; y += 8) {
285 memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
286 for (x = 0; x < width + 8; x += 8) {
287 int qp;
288
289 if (p->qp) {
290 qp = p->qp;
291 } else{
292 const int qps = 3 + is_luma;
293 qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
294 qp = FFMAX(1, ff_norm_qscale(qp, p->qscale_type));
295 }
296 for (i = 0; i < count; i++) {
297 const int x1 = x + offset[i + count - 1][0];
298 const int y1 = y + offset[i + count - 1][1];
299 const int index = x1 + y1*linesize;
300 p->dct->get_pixels_unaligned(block, p->src + sample_bytes*index, sample_bytes*linesize);
301 p->dct->fdct(block);
302 p->requantize(block2, block, qp, p->dct->idct_permutation);
303 p->dct->idct(block2);
304 add_block(p->temp + index, linesize, block2);
305 }
306 }
307 if (y) {
308 if (sample_bytes == 1) {
309 p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
310 dst_linesize, linesize, width,
311 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
312 ldither);
313 } else {
314 store_slice16_c((uint16_t*)(dst + (y - 8) * dst_linesize), p->temp + 8 + y*linesize,
315 dst_linesize/2, linesize, width,
316 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
317 ldither, depth);
318 }
319 }
320 }
321 }
322
query_formats(AVFilterContext * ctx)323 static int query_formats(AVFilterContext *ctx)
324 {
325 static const enum AVPixelFormat pix_fmts[] = {
326 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
327 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
328 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
329 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
330 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
331 AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
332 AV_PIX_FMT_YUV420P10,
333 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9,
334 AV_PIX_FMT_YUV420P9,
335 AV_PIX_FMT_GRAY8,
336 AV_PIX_FMT_GBRP,
337 AV_PIX_FMT_GBRP9,
338 AV_PIX_FMT_GBRP10,
339 AV_PIX_FMT_NONE
340 };
341
342 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
343 if (!fmts_list)
344 return AVERROR(ENOMEM);
345 return ff_set_common_formats(ctx, fmts_list);
346 }
347
config_input(AVFilterLink * inlink)348 static int config_input(AVFilterLink *inlink)
349 {
350 SPPContext *s = inlink->dst->priv;
351 const int h = FFALIGN(inlink->h + 16, 16);
352 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
353 const int bps = desc->comp[0].depth;
354
355 av_opt_set_int(s->dct, "bits_per_sample", bps, 0);
356 avcodec_dct_init(s->dct);
357
358 if (ARCH_X86)
359 ff_spp_init_x86(s);
360
361 s->hsub = desc->log2_chroma_w;
362 s->vsub = desc->log2_chroma_h;
363 s->temp_linesize = FFALIGN(inlink->w + 16, 16);
364 s->temp = av_malloc_array(s->temp_linesize, h * sizeof(*s->temp));
365 s->src = av_malloc_array(s->temp_linesize, h * sizeof(*s->src) * 2);
366
367 if (!s->temp || !s->src)
368 return AVERROR(ENOMEM);
369 return 0;
370 }
371
filter_frame(AVFilterLink * inlink,AVFrame * in)372 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
373 {
374 AVFilterContext *ctx = inlink->dst;
375 SPPContext *s = ctx->priv;
376 AVFilterLink *outlink = ctx->outputs[0];
377 AVFrame *out = in;
378 int qp_stride = 0;
379 int8_t *qp_table = NULL;
380 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
381 const int depth = desc->comp[0].depth;
382 int ret = 0;
383
384 /* if we are not in a constant user quantizer mode and we don't want to use
385 * the quantizers from the B-frames (B-frames often have a higher QP), we
386 * need to save the qp table from the last non B-frame; this is what the
387 * following code block does */
388 if (!s->qp && (s->use_bframe_qp || in->pict_type != AV_PICTURE_TYPE_B)) {
389 ret = ff_qp_table_extract(in, &qp_table, &qp_stride, NULL, &s->qscale_type);
390 if (ret < 0) {
391 av_frame_free(&in);
392 return ret;
393 }
394
395 if (!s->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
396 av_freep(&s->non_b_qp_table);
397 s->non_b_qp_table = qp_table;
398 s->non_b_qp_stride = qp_stride;
399 }
400 }
401
402 if (s->log2_count && !ctx->is_disabled) {
403 if (!s->use_bframe_qp && s->non_b_qp_table) {
404 qp_table = s->non_b_qp_table;
405 qp_stride = s->non_b_qp_stride;
406 }
407
408 if (qp_table || s->qp) {
409 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
410 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
411
412 /* get a new frame if in-place is not possible or if the dimensions
413 * are not multiple of 8 */
414 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
415 const int aligned_w = FFALIGN(inlink->w, 8);
416 const int aligned_h = FFALIGN(inlink->h, 8);
417
418 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
419 if (!out) {
420 av_frame_free(&in);
421 ret = AVERROR(ENOMEM);
422 goto finish;
423 }
424 av_frame_copy_props(out, in);
425 out->width = in->width;
426 out->height = in->height;
427 }
428
429 filter(s, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, depth);
430
431 if (out->data[2]) {
432 filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
433 filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
434 }
435 emms_c();
436 }
437 }
438
439 if (in != out) {
440 if (in->data[3])
441 av_image_copy_plane(out->data[3], out->linesize[3],
442 in ->data[3], in ->linesize[3],
443 inlink->w, inlink->h);
444 av_frame_free(&in);
445 }
446 ret = ff_filter_frame(outlink, out);
447 finish:
448 if (qp_table != s->non_b_qp_table)
449 av_freep(&qp_table);
450 return ret;
451 }
452
process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)453 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
454 char *res, int res_len, int flags)
455 {
456 SPPContext *s = ctx->priv;
457
458 if (!strcmp(cmd, "level") || !strcmp(cmd, "quality")) {
459 if (!strcmp(args, "max"))
460 s->log2_count = MAX_LEVEL;
461 else
462 s->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
463 return 0;
464 }
465 return AVERROR(ENOSYS);
466 }
467
init_dict(AVFilterContext * ctx,AVDictionary ** opts)468 static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
469 {
470 SPPContext *s = ctx->priv;
471 int ret;
472
473 s->dct = avcodec_dct_alloc();
474 if (!s->dct)
475 return AVERROR(ENOMEM);
476
477 if (opts) {
478 AVDictionaryEntry *e = NULL;
479
480 while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) {
481 if ((ret = av_opt_set(s->dct, e->key, e->value, 0)) < 0)
482 return ret;
483 }
484 av_dict_free(opts);
485 }
486
487 s->store_slice = store_slice_c;
488 switch (s->mode) {
489 case MODE_HARD: s->requantize = hardthresh_c; break;
490 case MODE_SOFT: s->requantize = softthresh_c; break;
491 }
492 return 0;
493 }
494
uninit(AVFilterContext * ctx)495 static av_cold void uninit(AVFilterContext *ctx)
496 {
497 SPPContext *s = ctx->priv;
498
499 av_freep(&s->temp);
500 av_freep(&s->src);
501 av_freep(&s->dct);
502 av_freep(&s->non_b_qp_table);
503 }
504
505 static const AVFilterPad spp_inputs[] = {
506 {
507 .name = "default",
508 .type = AVMEDIA_TYPE_VIDEO,
509 .config_props = config_input,
510 .filter_frame = filter_frame,
511 },
512 { NULL }
513 };
514
515 static const AVFilterPad spp_outputs[] = {
516 {
517 .name = "default",
518 .type = AVMEDIA_TYPE_VIDEO,
519 },
520 { NULL }
521 };
522
523 AVFilter ff_vf_spp = {
524 .name = "spp",
525 .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."),
526 .priv_size = sizeof(SPPContext),
527 .init_dict = init_dict,
528 .uninit = uninit,
529 .query_formats = query_formats,
530 .inputs = spp_inputs,
531 .outputs = spp_outputs,
532 .process_command = process_command,
533 .priv_class = &spp_class,
534 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
535 };
536