2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Clément Bœsch <u pkh me>
5 * This file is part of FFmpeg.
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.
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.
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.
24 * Simple post processing filter
26 * This implementation is based on an algorithm described in
27 * "Aria Nosratinia Embedded Post-Processing for
28 * Enhancement of Compressed Images (1999)"
30 * Originally written by Michael Niedermayer for the MPlayer project, and
31 * ported by Clément Bœsch for FFmpeg.
34 #include "libavutil/avassert.h"
35 #include "libavutil/imgutils.h"
36 #include "libavutil/opt.h"
37 #include "libavutil/pixdesc.h"
47 #if FF_API_CHILD_CLASS_NEXT
48 static const AVClass *child_class_next(const AVClass *prev)
50 return prev ? NULL : avcodec_dct_get_class();
54 static const AVClass *child_class_iterate(void **iter)
56 const AVClass *c = *iter ? NULL : avcodec_dct_get_class();
57 *iter = (void*)(uintptr_t)c;
61 static void *child_next(void *obj, void *prev)
64 return prev ? NULL : s->dct;
67 #define OFFSET(x) offsetof(SPPContext, x)
68 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
69 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
70 static const AVOption spp_options[] = {
71 { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, TFLAGS },
72 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
73 { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
74 { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
75 { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
76 { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS },
80 static const AVClass spp_class = {
82 .item_name = av_default_item_name,
83 .option = spp_options,
84 .version = LIBAVUTIL_VERSION_INT,
85 .category = AV_CLASS_CATEGORY_FILTER,
86 #if FF_API_CHILD_CLASS_NEXT
87 .child_class_next = child_class_next,
89 .child_class_iterate = child_class_iterate,
90 .child_next = child_next,
93 // XXX: share between filters?
94 DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
95 { 0, 48, 12, 60, 3, 51, 15, 63 },
96 { 32, 16, 44, 28, 35, 19, 47, 31 },
97 { 8, 56, 4, 52, 11, 59, 7, 55 },
98 { 40, 24, 36, 20, 43, 27, 39, 23 },
99 { 2, 50, 14, 62, 1, 49, 13, 61 },
100 { 34, 18, 46, 30, 33, 17, 45, 29 },
101 { 10, 58, 6, 54, 9, 57, 5, 53 },
102 { 42, 26, 38, 22, 41, 25, 37, 21 },
105 static const uint8_t offset[127][2] = {
107 {0,0}, {4,4}, // quality = 1
108 {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2
109 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
111 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
112 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
114 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
115 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
116 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
117 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
119 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
120 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
121 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
122 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
123 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
124 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
125 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
126 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
129 static void hardthresh_c(int16_t dst[64], const int16_t src[64],
130 int qp, const uint8_t *permutation)
133 int bias = 0; // FIXME
135 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
136 unsigned threshold2 = threshold1 << 1;
138 memset(dst, 0, 64 * sizeof(dst[0]));
139 dst[0] = (src[0] + 4) >> 3;
141 for (i = 1; i < 64; i++) {
143 if (((unsigned)(level + threshold1)) > threshold2) {
144 const int j = permutation[i];
145 dst[j] = (level + 4) >> 3;
150 static void softthresh_c(int16_t dst[64], const int16_t src[64],
151 int qp, const uint8_t *permutation)
154 int bias = 0; //FIXME
156 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
157 unsigned threshold2 = threshold1 << 1;
159 memset(dst, 0, 64 * sizeof(dst[0]));
160 dst[0] = (src[0] + 4) >> 3;
162 for (i = 1; i < 64; i++) {
164 if (((unsigned)(level + threshold1)) > threshold2) {
165 const int j = permutation[i];
166 if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
167 else dst[j] = (level + threshold1 + 4) >> 3;
172 static void store_slice_c(uint8_t *dst, const int16_t *src,
173 int dst_linesize, int src_linesize,
174 int width, int height, int log2_scale,
175 const uint8_t dither[8][8])
179 #define STORE(pos) do { \
180 temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \
182 temp = ~(temp >> 31); \
183 dst[x + y*dst_linesize + pos] = temp; \
186 for (y = 0; y < height; y++) {
187 const uint8_t *d = dither[y];
188 for (x = 0; x < width; x += 8) {
202 static void store_slice16_c(uint16_t *dst, const int16_t *src,
203 int dst_linesize, int src_linesize,
204 int width, int height, int log2_scale,
205 const uint8_t dither[8][8], int depth)
208 unsigned int mask = -1<<depth;
210 #define STORE16(pos) do { \
211 temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5; \
213 temp = ~(temp >> 31); \
214 dst[x + y*dst_linesize + pos] = temp; \
217 for (y = 0; y < height; y++) {
218 const uint8_t *d = dither[y];
219 for (x = 0; x < width; x += 8) {
233 static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
237 for (y = 0; y < 8; y++) {
238 dst[0 + y*linesize] += block[0 + y*8];
239 dst[1 + y*linesize] += block[1 + y*8];
240 dst[2 + y*linesize] += block[2 + y*8];
241 dst[3 + y*linesize] += block[3 + y*8];
242 dst[4 + y*linesize] += block[4 + y*8];
243 dst[5 + y*linesize] += block[5 + y*8];
244 dst[6 + y*linesize] += block[6 + y*8];
245 dst[7 + y*linesize] += block[7 + y*8];
249 static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
250 int dst_linesize, int src_linesize, int width, int height,
251 const uint8_t *qp_table, int qp_stride, int is_luma, int depth)
254 const int count = 1 << p->log2_count;
255 const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
256 DECLARE_ALIGNED(16, uint64_t, block_align)[32];
257 int16_t *block = (int16_t *)block_align;
258 int16_t *block2 = (int16_t *)(block_align + 16);
259 uint16_t *psrc16 = (uint16_t*)p->src;
260 const int sample_bytes = (depth+7) / 8;
262 for (y = 0; y < height; y++) {
263 int index = 8 + 8*linesize + y*linesize;
264 memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
265 if (sample_bytes == 1) {
266 for (x = 0; x < 8; x++) {
267 p->src[index - x - 1] = p->src[index + x ];
268 p->src[index + width + x ] = p->src[index + width - x - 1];
271 for (x = 0; x < 8; x++) {
272 psrc16[index - x - 1] = psrc16[index + x ];
273 psrc16[index + width + x ] = psrc16[index + width - x - 1];
277 for (y = 0; y < 8; y++) {
278 memcpy(p->src + ( 7-y)*linesize * sample_bytes, p->src + ( y+8)*linesize * sample_bytes, linesize * sample_bytes);
279 memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
282 for (y = 0; y < height + 8; y += 8) {
283 memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
284 for (x = 0; x < width + 8; x += 8) {
290 const int qps = 3 + is_luma;
291 qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
292 qp = FFMAX(1, ff_norm_qscale(qp, p->qscale_type));
294 for (i = 0; i < count; i++) {
295 const int x1 = x + offset[i + count - 1][0];
296 const int y1 = y + offset[i + count - 1][1];
297 const int index = x1 + y1*linesize;
298 p->dct->get_pixels_unaligned(block, p->src + sample_bytes*index, sample_bytes*linesize);
300 p->requantize(block2, block, qp, p->dct->idct_permutation);
301 p->dct->idct(block2);
302 add_block(p->temp + index, linesize, block2);
306 if (sample_bytes == 1) {
307 p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
308 dst_linesize, linesize, width,
309 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
312 store_slice16_c((uint16_t*)(dst + (y - 8) * dst_linesize), p->temp + 8 + y*linesize,
313 dst_linesize/2, linesize, width,
314 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
321 static int query_formats(AVFilterContext *ctx)
323 static const enum AVPixelFormat pix_fmts[] = {
324 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
325 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
326 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
327 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
328 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
329 AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
330 AV_PIX_FMT_YUV420P10,
331 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9,
340 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
342 return AVERROR(ENOMEM);
343 return ff_set_common_formats(ctx, fmts_list);
346 static int config_input(AVFilterLink *inlink)
348 SPPContext *s = inlink->dst->priv;
349 const int h = FFALIGN(inlink->h + 16, 16);
350 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
351 const int bps = desc->comp[0].depth;
353 av_opt_set_int(s->dct, "bits_per_sample", bps, 0);
354 avcodec_dct_init(s->dct);
359 s->hsub = desc->log2_chroma_w;
360 s->vsub = desc->log2_chroma_h;
361 s->temp_linesize = FFALIGN(inlink->w + 16, 16);
362 s->temp = av_malloc_array(s->temp_linesize, h * sizeof(*s->temp));
363 s->src = av_malloc_array(s->temp_linesize, h * sizeof(*s->src) * 2);
365 if (!s->temp || !s->src)
366 return AVERROR(ENOMEM);
370 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
372 AVFilterContext *ctx = inlink->dst;
373 SPPContext *s = ctx->priv;
374 AVFilterLink *outlink = ctx->outputs[0];
377 const int8_t *qp_table = NULL;
378 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
379 const int depth = desc->comp[0].depth;
381 /* if we are not in a constant user quantizer mode and we don't want to use
382 * the quantizers from the B-frames (B-frames often have a higher QP), we
383 * need to save the qp table from the last non B-frame; this is what the
384 * following code block does */
386 qp_table = av_frame_get_qp_table(in, &qp_stride, &s->qscale_type);
388 if (qp_table && !s->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
391 /* if the qp stride is not set, it means the QP are only defined on
394 w = AV_CEIL_RSHIFT(inlink->w, 4);
398 h = AV_CEIL_RSHIFT(inlink->h, 4);
401 if (w * h > s->non_b_qp_alloc_size) {
402 int ret = av_reallocp_array(&s->non_b_qp_table, w, h);
404 s->non_b_qp_alloc_size = 0;
407 s->non_b_qp_alloc_size = w * h;
410 av_assert0(w * h <= s->non_b_qp_alloc_size);
411 memcpy(s->non_b_qp_table, qp_table, w * h);
415 if (s->log2_count && !ctx->is_disabled) {
416 if (!s->use_bframe_qp && s->non_b_qp_table)
417 qp_table = s->non_b_qp_table;
419 if (qp_table || s->qp) {
420 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
421 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
423 /* get a new frame if in-place is not possible or if the dimensions
424 * are not multiple of 8 */
425 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
426 const int aligned_w = FFALIGN(inlink->w, 8);
427 const int aligned_h = FFALIGN(inlink->h, 8);
429 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
432 return AVERROR(ENOMEM);
434 av_frame_copy_props(out, in);
435 out->width = in->width;
436 out->height = in->height;
439 filter(s, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, depth);
442 filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
443 filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
451 av_image_copy_plane(out->data[3], out->linesize[3],
452 in ->data[3], in ->linesize[3],
453 inlink->w, inlink->h);
456 return ff_filter_frame(outlink, out);
459 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
460 char *res, int res_len, int flags)
462 SPPContext *s = ctx->priv;
464 if (!strcmp(cmd, "level") || !strcmp(cmd, "quality")) {
465 if (!strcmp(args, "max"))
466 s->log2_count = MAX_LEVEL;
468 s->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
471 return AVERROR(ENOSYS);
474 static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
476 SPPContext *s = ctx->priv;
479 s->dct = avcodec_dct_alloc();
481 return AVERROR(ENOMEM);
484 AVDictionaryEntry *e = NULL;
486 while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) {
487 if ((ret = av_opt_set(s->dct, e->key, e->value, 0)) < 0)
493 s->store_slice = store_slice_c;
495 case MODE_HARD: s->requantize = hardthresh_c; break;
496 case MODE_SOFT: s->requantize = softthresh_c; break;
501 static av_cold void uninit(AVFilterContext *ctx)
503 SPPContext *s = ctx->priv;
508 av_freep(&s->non_b_qp_table);
511 static const AVFilterPad spp_inputs[] = {
514 .type = AVMEDIA_TYPE_VIDEO,
515 .config_props = config_input,
516 .filter_frame = filter_frame,
521 static const AVFilterPad spp_outputs[] = {
524 .type = AVMEDIA_TYPE_VIDEO,
529 AVFilter ff_vf_spp = {
531 .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."),
532 .priv_size = sizeof(SPPContext),
533 .init_dict = init_dict,
535 .query_formats = query_formats,
536 .inputs = spp_inputs,
537 .outputs = spp_outputs,
538 .process_command = process_command,
539 .priv_class = &spp_class,
540 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,