Commit | Line | Data |
---|---|---|
2ba45a60 DM |
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/opt.h" | |
37 | #include "libavutil/pixdesc.h" | |
38 | #include "internal.h" | |
39 | #include "vf_spp.h" | |
40 | ||
41 | enum mode { | |
42 | MODE_HARD, | |
43 | MODE_SOFT, | |
44 | NB_MODES | |
45 | }; | |
46 | ||
47 | static const AVClass *child_class_next(const AVClass *prev) | |
48 | { | |
49 | return prev ? NULL : avcodec_dct_get_class(); | |
50 | } | |
51 | ||
52 | static void *child_next(void *obj, void *prev) | |
53 | { | |
54 | SPPContext *s = obj; | |
55 | return prev ? NULL : s->dct; | |
56 | } | |
57 | ||
58 | #define OFFSET(x) offsetof(SPPContext, x) | |
59 | #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM | |
60 | static const AVOption spp_options[] = { | |
61 | { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, FLAGS }, | |
62 | { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS }, | |
63 | { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" }, | |
64 | { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" }, | |
65 | { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" }, | |
66 | { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS }, | |
67 | { NULL } | |
68 | }; | |
69 | ||
70 | static const AVClass spp_class = { | |
71 | .class_name = "spp", | |
72 | .item_name = av_default_item_name, | |
73 | .option = spp_options, | |
74 | .version = LIBAVUTIL_VERSION_INT, | |
75 | .category = AV_CLASS_CATEGORY_FILTER, | |
76 | .child_class_next = child_class_next, | |
77 | .child_next = child_next, | |
78 | }; | |
79 | ||
80 | // XXX: share between filters? | |
81 | DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = { | |
82 | { 0, 48, 12, 60, 3, 51, 15, 63 }, | |
83 | { 32, 16, 44, 28, 35, 19, 47, 31 }, | |
84 | { 8, 56, 4, 52, 11, 59, 7, 55 }, | |
85 | { 40, 24, 36, 20, 43, 27, 39, 23 }, | |
86 | { 2, 50, 14, 62, 1, 49, 13, 61 }, | |
87 | { 34, 18, 46, 30, 33, 17, 45, 29 }, | |
88 | { 10, 58, 6, 54, 9, 57, 5, 53 }, | |
89 | { 42, 26, 38, 22, 41, 25, 37, 21 }, | |
90 | }; | |
91 | ||
92 | static const uint8_t offset[127][2] = { | |
93 | {0,0}, | |
94 | {0,0}, {4,4}, // quality = 1 | |
95 | {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2 | |
96 | {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3 | |
97 | ||
98 | {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4 | |
99 | {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7}, | |
100 | ||
101 | {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5 | |
102 | {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7}, | |
103 | {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7}, | |
104 | {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7}, | |
105 | ||
106 | {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6 | |
107 | {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0}, | |
108 | {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3}, | |
109 | {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1}, | |
110 | {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3}, | |
111 | {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1}, | |
112 | {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2}, | |
113 | {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0}, | |
114 | }; | |
115 | ||
116 | static void hardthresh_c(int16_t dst[64], const int16_t src[64], | |
117 | int qp, const uint8_t *permutation) | |
118 | { | |
119 | int i; | |
120 | int bias = 0; // FIXME | |
121 | ||
122 | unsigned threshold1 = qp * ((1<<4) - bias) - 1; | |
123 | unsigned threshold2 = threshold1 << 1; | |
124 | ||
125 | memset(dst, 0, 64 * sizeof(dst[0])); | |
126 | dst[0] = (src[0] + 4) >> 3; | |
127 | ||
128 | for (i = 1; i < 64; i++) { | |
129 | int level = src[i]; | |
130 | if (((unsigned)(level + threshold1)) > threshold2) { | |
131 | const int j = permutation[i]; | |
132 | dst[j] = (level + 4) >> 3; | |
133 | } | |
134 | } | |
135 | } | |
136 | ||
137 | static void softthresh_c(int16_t dst[64], const int16_t src[64], | |
138 | int qp, const uint8_t *permutation) | |
139 | { | |
140 | int i; | |
141 | int bias = 0; //FIXME | |
142 | ||
143 | unsigned threshold1 = qp * ((1<<4) - bias) - 1; | |
144 | unsigned threshold2 = threshold1 << 1; | |
145 | ||
146 | memset(dst, 0, 64 * sizeof(dst[0])); | |
147 | dst[0] = (src[0] + 4) >> 3; | |
148 | ||
149 | for (i = 1; i < 64; i++) { | |
150 | int level = src[i]; | |
151 | if (((unsigned)(level + threshold1)) > threshold2) { | |
152 | const int j = permutation[i]; | |
153 | if (level > 0) dst[j] = (level - threshold1 + 4) >> 3; | |
154 | else dst[j] = (level + threshold1 + 4) >> 3; | |
155 | } | |
156 | } | |
157 | } | |
158 | ||
159 | static void store_slice_c(uint8_t *dst, const int16_t *src, | |
160 | int dst_linesize, int src_linesize, | |
161 | int width, int height, int log2_scale, | |
162 | const uint8_t dither[8][8]) | |
163 | { | |
164 | int y, x; | |
165 | ||
166 | #define STORE(pos) do { \ | |
167 | temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \ | |
168 | if (temp & 0x100) \ | |
169 | temp = ~(temp >> 31); \ | |
170 | dst[x + y*dst_linesize + pos] = temp; \ | |
171 | } while (0) | |
172 | ||
173 | for (y = 0; y < height; y++) { | |
174 | const uint8_t *d = dither[y]; | |
175 | for (x = 0; x < width; x += 8) { | |
176 | int temp; | |
177 | STORE(0); | |
178 | STORE(1); | |
179 | STORE(2); | |
180 | STORE(3); | |
181 | STORE(4); | |
182 | STORE(5); | |
183 | STORE(6); | |
184 | STORE(7); | |
185 | } | |
186 | } | |
187 | } | |
188 | ||
189 | static inline void add_block(int16_t *dst, int linesize, const int16_t block[64]) | |
190 | { | |
191 | int y; | |
192 | ||
193 | for (y = 0; y < 8; y++) { | |
194 | *(uint32_t *)&dst[0 + y*linesize] += *(uint32_t *)&block[0 + y*8]; | |
195 | *(uint32_t *)&dst[2 + y*linesize] += *(uint32_t *)&block[2 + y*8]; | |
196 | *(uint32_t *)&dst[4 + y*linesize] += *(uint32_t *)&block[4 + y*8]; | |
197 | *(uint32_t *)&dst[6 + y*linesize] += *(uint32_t *)&block[6 + y*8]; | |
198 | } | |
199 | } | |
200 | ||
201 | // XXX: export the function? | |
202 | static inline int norm_qscale(int qscale, int type) | |
203 | { | |
204 | switch (type) { | |
205 | case FF_QSCALE_TYPE_MPEG1: return qscale; | |
206 | case FF_QSCALE_TYPE_MPEG2: return qscale >> 1; | |
207 | case FF_QSCALE_TYPE_H264: return qscale >> 2; | |
208 | case FF_QSCALE_TYPE_VP56: return (63 - qscale + 2) >> 2; | |
209 | } | |
210 | return qscale; | |
211 | } | |
212 | ||
213 | static void filter(SPPContext *p, uint8_t *dst, uint8_t *src, | |
214 | int dst_linesize, int src_linesize, int width, int height, | |
215 | const uint8_t *qp_table, int qp_stride, int is_luma) | |
216 | { | |
217 | int x, y, i; | |
218 | const int count = 1 << p->log2_count; | |
219 | const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16); | |
220 | DECLARE_ALIGNED(16, uint64_t, block_align)[32]; | |
221 | int16_t *block = (int16_t *)block_align; | |
222 | int16_t *block2 = (int16_t *)(block_align + 16); | |
223 | ||
224 | for (y = 0; y < height; y++) { | |
225 | int index = 8 + 8*linesize + y*linesize; | |
226 | memcpy(p->src + index, src + y*src_linesize, width); | |
227 | for (x = 0; x < 8; x++) { | |
228 | p->src[index - x - 1] = p->src[index + x ]; | |
229 | p->src[index + width + x ] = p->src[index + width - x - 1]; | |
230 | } | |
231 | } | |
232 | for (y = 0; y < 8; y++) { | |
233 | memcpy(p->src + ( 7-y)*linesize, p->src + ( y+8)*linesize, linesize); | |
234 | memcpy(p->src + (height+8+y)*linesize, p->src + (height-y+7)*linesize, linesize); | |
235 | } | |
236 | ||
237 | for (y = 0; y < height + 8; y += 8) { | |
238 | memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp)); | |
239 | for (x = 0; x < width + 8; x += 8) { | |
240 | int qp; | |
241 | ||
242 | if (p->qp) { | |
243 | qp = p->qp; | |
244 | } else{ | |
245 | const int qps = 3 + is_luma; | |
246 | qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride]; | |
247 | qp = FFMAX(1, norm_qscale(qp, p->qscale_type)); | |
248 | } | |
249 | for (i = 0; i < count; i++) { | |
250 | const int x1 = x + offset[i + count - 1][0]; | |
251 | const int y1 = y + offset[i + count - 1][1]; | |
252 | const int index = x1 + y1*linesize; | |
253 | p->dct->get_pixels(block, p->src + index, linesize); | |
254 | p->dct->fdct(block); | |
255 | p->requantize(block2, block, qp, p->dct->idct_permutation); | |
256 | p->dct->idct(block2); | |
257 | add_block(p->temp + index, linesize, block2); | |
258 | } | |
259 | } | |
260 | if (y) | |
261 | p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize, | |
262 | dst_linesize, linesize, width, | |
263 | FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count, | |
264 | ldither); | |
265 | } | |
266 | } | |
267 | ||
268 | static int query_formats(AVFilterContext *ctx) | |
269 | { | |
270 | static const enum PixelFormat pix_fmts[] = { | |
271 | AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, | |
272 | AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P, | |
273 | AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P, | |
274 | AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P, | |
275 | AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P, | |
276 | AV_PIX_FMT_NONE | |
277 | }; | |
278 | ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); | |
279 | return 0; | |
280 | } | |
281 | ||
282 | static int config_input(AVFilterLink *inlink) | |
283 | { | |
284 | SPPContext *spp = inlink->dst->priv; | |
285 | const int h = FFALIGN(inlink->h + 16, 16); | |
286 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); | |
287 | ||
288 | spp->hsub = desc->log2_chroma_w; | |
289 | spp->vsub = desc->log2_chroma_h; | |
290 | spp->temp_linesize = FFALIGN(inlink->w + 16, 16); | |
291 | spp->temp = av_malloc_array(spp->temp_linesize, h * sizeof(*spp->temp)); | |
292 | spp->src = av_malloc_array(spp->temp_linesize, h * sizeof(*spp->src)); | |
293 | if (!spp->use_bframe_qp) { | |
294 | /* we are assuming here the qp blocks will not be smaller that 16x16 */ | |
295 | spp->non_b_qp_alloc_size = FF_CEIL_RSHIFT(inlink->w, 4) * FF_CEIL_RSHIFT(inlink->h, 4); | |
296 | spp->non_b_qp_table = av_calloc(spp->non_b_qp_alloc_size, sizeof(*spp->non_b_qp_table)); | |
297 | if (!spp->non_b_qp_table) | |
298 | return AVERROR(ENOMEM); | |
299 | } | |
300 | if (!spp->temp || !spp->src) | |
301 | return AVERROR(ENOMEM); | |
302 | return 0; | |
303 | } | |
304 | ||
305 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) | |
306 | { | |
307 | AVFilterContext *ctx = inlink->dst; | |
308 | SPPContext *spp = ctx->priv; | |
309 | AVFilterLink *outlink = ctx->outputs[0]; | |
310 | AVFrame *out = in; | |
311 | int qp_stride = 0; | |
312 | const int8_t *qp_table = NULL; | |
313 | ||
314 | /* if we are not in a constant user quantizer mode and we don't want to use | |
315 | * the quantizers from the B-frames (B-frames often have a higher QP), we | |
316 | * need to save the qp table from the last non B-frame; this is what the | |
317 | * following code block does */ | |
318 | if (!spp->qp) { | |
319 | qp_table = av_frame_get_qp_table(in, &qp_stride, &spp->qscale_type); | |
320 | ||
321 | if (qp_table && !spp->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) { | |
322 | int w, h; | |
323 | ||
324 | /* if the qp stride is not set, it means the QP are only defined on | |
325 | * a line basis */ | |
326 | if (!qp_stride) { | |
327 | w = FF_CEIL_RSHIFT(inlink->w, 4); | |
328 | h = 1; | |
329 | } else { | |
330 | w = FF_CEIL_RSHIFT(qp_stride, 4); | |
331 | h = FF_CEIL_RSHIFT(inlink->h, 4); | |
332 | } | |
333 | av_assert0(w * h <= spp->non_b_qp_alloc_size); | |
334 | memcpy(spp->non_b_qp_table, qp_table, w * h); | |
335 | } | |
336 | } | |
337 | ||
338 | if (spp->log2_count && !ctx->is_disabled) { | |
339 | if (!spp->use_bframe_qp && spp->non_b_qp_table) | |
340 | qp_table = spp->non_b_qp_table; | |
341 | ||
342 | if (qp_table || spp->qp) { | |
343 | const int cw = FF_CEIL_RSHIFT(inlink->w, spp->hsub); | |
344 | const int ch = FF_CEIL_RSHIFT(inlink->h, spp->vsub); | |
345 | ||
346 | /* get a new frame if in-place is not possible or if the dimensions | |
347 | * are not multiple of 8 */ | |
348 | if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) { | |
349 | const int aligned_w = FFALIGN(inlink->w, 8); | |
350 | const int aligned_h = FFALIGN(inlink->h, 8); | |
351 | ||
352 | out = ff_get_video_buffer(outlink, aligned_w, aligned_h); | |
353 | if (!out) { | |
354 | av_frame_free(&in); | |
355 | return AVERROR(ENOMEM); | |
356 | } | |
357 | av_frame_copy_props(out, in); | |
358 | out->width = in->width; | |
359 | out->height = in->height; | |
360 | } | |
361 | ||
362 | filter(spp, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1); | |
363 | filter(spp, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0); | |
364 | filter(spp, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0); | |
365 | emms_c(); | |
366 | } | |
367 | } | |
368 | ||
369 | if (in != out) { | |
370 | if (in->data[3]) | |
371 | av_image_copy_plane(out->data[3], out->linesize[3], | |
372 | in ->data[3], in ->linesize[3], | |
373 | inlink->w, inlink->h); | |
374 | av_frame_free(&in); | |
375 | } | |
376 | return ff_filter_frame(outlink, out); | |
377 | } | |
378 | ||
379 | static int process_command(AVFilterContext *ctx, const char *cmd, const char *args, | |
380 | char *res, int res_len, int flags) | |
381 | { | |
382 | SPPContext *spp = ctx->priv; | |
383 | ||
384 | if (!strcmp(cmd, "level")) { | |
385 | if (!strcmp(args, "max")) | |
386 | spp->log2_count = MAX_LEVEL; | |
387 | else | |
388 | spp->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL); | |
389 | return 0; | |
390 | } | |
391 | return AVERROR(ENOSYS); | |
392 | } | |
393 | ||
394 | static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts) | |
395 | { | |
396 | SPPContext *spp = ctx->priv; | |
397 | int ret; | |
398 | ||
399 | spp->avctx = avcodec_alloc_context3(NULL); | |
400 | spp->dct = avcodec_dct_alloc(); | |
401 | if (!spp->avctx || !spp->dct) | |
402 | return AVERROR(ENOMEM); | |
403 | ||
404 | if (opts) { | |
405 | AVDictionaryEntry *e = NULL; | |
406 | ||
407 | while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) { | |
408 | if ((ret = av_opt_set(spp->dct, e->key, e->value, 0)) < 0) | |
409 | return ret; | |
410 | } | |
411 | av_dict_free(opts); | |
412 | } | |
413 | ||
414 | avcodec_dct_init(spp->dct); | |
415 | spp->store_slice = store_slice_c; | |
416 | switch (spp->mode) { | |
417 | case MODE_HARD: spp->requantize = hardthresh_c; break; | |
418 | case MODE_SOFT: spp->requantize = softthresh_c; break; | |
419 | } | |
420 | if (ARCH_X86) | |
421 | ff_spp_init_x86(spp); | |
422 | return 0; | |
423 | } | |
424 | ||
425 | static av_cold void uninit(AVFilterContext *ctx) | |
426 | { | |
427 | SPPContext *spp = ctx->priv; | |
428 | ||
429 | av_freep(&spp->temp); | |
430 | av_freep(&spp->src); | |
431 | if (spp->avctx) { | |
432 | avcodec_close(spp->avctx); | |
433 | av_freep(&spp->avctx); | |
434 | } | |
435 | av_freep(&spp->dct); | |
436 | av_freep(&spp->non_b_qp_table); | |
437 | } | |
438 | ||
439 | static const AVFilterPad spp_inputs[] = { | |
440 | { | |
441 | .name = "default", | |
442 | .type = AVMEDIA_TYPE_VIDEO, | |
443 | .config_props = config_input, | |
444 | .filter_frame = filter_frame, | |
445 | }, | |
446 | { NULL } | |
447 | }; | |
448 | ||
449 | static const AVFilterPad spp_outputs[] = { | |
450 | { | |
451 | .name = "default", | |
452 | .type = AVMEDIA_TYPE_VIDEO, | |
453 | }, | |
454 | { NULL } | |
455 | }; | |
456 | ||
457 | AVFilter ff_vf_spp = { | |
458 | .name = "spp", | |
459 | .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."), | |
460 | .priv_size = sizeof(SPPContext), | |
461 | .init_dict = init_dict, | |
462 | .uninit = uninit, | |
463 | .query_formats = query_formats, | |
464 | .inputs = spp_inputs, | |
465 | .outputs = spp_outputs, | |
466 | .process_command = process_command, | |
467 | .priv_class = &spp_class, | |
468 | .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, | |
469 | }; |