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