You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

531 lines
18KB

  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. * @file
  23. * Simple post processing filter
  24. *
  25. * This implementation is based on an algorithm described in
  26. * "Aria Nosratinia Embedded Post-Processing for
  27. * Enhancement of Compressed Images (1999)"
  28. *
  29. * Originally written by Michael Niedermayer for the MPlayer project, and
  30. * ported by Clément Bœsch for FFmpeg.
  31. */
  32. #include "libavutil/avassert.h"
  33. #include "libavutil/imgutils.h"
  34. #include "libavutil/opt.h"
  35. #include "libavutil/pixdesc.h"
  36. #include "internal.h"
  37. #include "vf_spp.h"
  38. enum mode {
  39. MODE_HARD,
  40. MODE_SOFT,
  41. NB_MODES
  42. };
  43. static const AVClass *child_class_next(const AVClass *prev)
  44. {
  45. return prev ? NULL : avcodec_dct_get_class();
  46. }
  47. static void *child_next(void *obj, void *prev)
  48. {
  49. SPPContext *s = obj;
  50. return prev ? NULL : s->dct;
  51. }
  52. #define OFFSET(x) offsetof(SPPContext, x)
  53. #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
  54. #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
  55. static const AVOption spp_options[] = {
  56. { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, TFLAGS },
  57. { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
  58. { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
  59. { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
  60. { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
  61. { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS },
  62. { NULL }
  63. };
  64. static const AVClass spp_class = {
  65. .class_name = "spp",
  66. .item_name = av_default_item_name,
  67. .option = spp_options,
  68. .version = LIBAVUTIL_VERSION_INT,
  69. .category = AV_CLASS_CATEGORY_FILTER,
  70. .child_class_next = child_class_next,
  71. .child_next = child_next,
  72. };
  73. // XXX: share between filters?
  74. DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
  75. { 0, 48, 12, 60, 3, 51, 15, 63 },
  76. { 32, 16, 44, 28, 35, 19, 47, 31 },
  77. { 8, 56, 4, 52, 11, 59, 7, 55 },
  78. { 40, 24, 36, 20, 43, 27, 39, 23 },
  79. { 2, 50, 14, 62, 1, 49, 13, 61 },
  80. { 34, 18, 46, 30, 33, 17, 45, 29 },
  81. { 10, 58, 6, 54, 9, 57, 5, 53 },
  82. { 42, 26, 38, 22, 41, 25, 37, 21 },
  83. };
  84. static const uint8_t offset[127][2] = {
  85. {0,0},
  86. {0,0}, {4,4}, // quality = 1
  87. {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2
  88. {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
  89. {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
  90. {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
  91. {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
  92. {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
  93. {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
  94. {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
  95. {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
  96. {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
  97. {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
  98. {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
  99. {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
  100. {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
  101. {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
  102. {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
  103. };
  104. static void hardthresh_c(int16_t dst[64], const int16_t src[64],
  105. int qp, const uint8_t *permutation)
  106. {
  107. int i;
  108. int bias = 0; // FIXME
  109. unsigned threshold1 = qp * ((1<<4) - bias) - 1;
  110. unsigned threshold2 = threshold1 << 1;
  111. memset(dst, 0, 64 * sizeof(dst[0]));
  112. dst[0] = (src[0] + 4) >> 3;
  113. for (i = 1; i < 64; i++) {
  114. int level = src[i];
  115. if (((unsigned)(level + threshold1)) > threshold2) {
  116. const int j = permutation[i];
  117. dst[j] = (level + 4) >> 3;
  118. }
  119. }
  120. }
  121. static void softthresh_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. unsigned threshold1 = qp * ((1<<4) - bias) - 1;
  127. unsigned threshold2 = threshold1 << 1;
  128. memset(dst, 0, 64 * sizeof(dst[0]));
  129. dst[0] = (src[0] + 4) >> 3;
  130. for (i = 1; i < 64; i++) {
  131. int level = src[i];
  132. if (((unsigned)(level + threshold1)) > threshold2) {
  133. const int j = permutation[i];
  134. if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
  135. else dst[j] = (level + threshold1 + 4) >> 3;
  136. }
  137. }
  138. }
  139. static void store_slice_c(uint8_t *dst, const int16_t *src,
  140. int dst_linesize, int src_linesize,
  141. int width, int height, int log2_scale,
  142. const uint8_t dither[8][8])
  143. {
  144. int y, x;
  145. #define STORE(pos) do { \
  146. temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \
  147. if (temp & 0x100) \
  148. temp = ~(temp >> 31); \
  149. dst[x + y*dst_linesize + pos] = temp; \
  150. } while (0)
  151. for (y = 0; y < height; y++) {
  152. const uint8_t *d = dither[y];
  153. for (x = 0; x < width; x += 8) {
  154. int temp;
  155. STORE(0);
  156. STORE(1);
  157. STORE(2);
  158. STORE(3);
  159. STORE(4);
  160. STORE(5);
  161. STORE(6);
  162. STORE(7);
  163. }
  164. }
  165. }
  166. static void store_slice16_c(uint16_t *dst, const int16_t *src,
  167. int dst_linesize, int src_linesize,
  168. int width, int height, int log2_scale,
  169. const uint8_t dither[8][8], int depth)
  170. {
  171. int y, x;
  172. unsigned int mask = -1<<depth;
  173. #define STORE16(pos) do { \
  174. temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5; \
  175. if (temp & mask ) \
  176. temp = ~(temp >> 31); \
  177. dst[x + y*dst_linesize + pos] = temp; \
  178. } while (0)
  179. for (y = 0; y < height; y++) {
  180. const uint8_t *d = dither[y];
  181. for (x = 0; x < width; x += 8) {
  182. int temp;
  183. STORE16(0);
  184. STORE16(1);
  185. STORE16(2);
  186. STORE16(3);
  187. STORE16(4);
  188. STORE16(5);
  189. STORE16(6);
  190. STORE16(7);
  191. }
  192. }
  193. }
  194. static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
  195. {
  196. int y;
  197. for (y = 0; y < 8; y++) {
  198. *(uint32_t *)&dst[0 + y*linesize] += *(uint32_t *)&block[0 + y*8];
  199. *(uint32_t *)&dst[2 + y*linesize] += *(uint32_t *)&block[2 + y*8];
  200. *(uint32_t *)&dst[4 + y*linesize] += *(uint32_t *)&block[4 + y*8];
  201. *(uint32_t *)&dst[6 + y*linesize] += *(uint32_t *)&block[6 + y*8];
  202. }
  203. }
  204. static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
  205. int dst_linesize, int src_linesize, int width, int height,
  206. const uint8_t *qp_table, int qp_stride, int is_luma, int depth)
  207. {
  208. int x, y, i;
  209. const int count = 1 << p->log2_count;
  210. const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
  211. DECLARE_ALIGNED(16, uint64_t, block_align)[32];
  212. int16_t *block = (int16_t *)block_align;
  213. int16_t *block2 = (int16_t *)(block_align + 16);
  214. uint16_t *psrc16 = (uint16_t*)p->src;
  215. const int sample_bytes = (depth+7) / 8;
  216. for (y = 0; y < height; y++) {
  217. int index = 8 + 8*linesize + y*linesize;
  218. memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
  219. if (sample_bytes == 1) {
  220. for (x = 0; x < 8; x++) {
  221. p->src[index - x - 1] = p->src[index + x ];
  222. p->src[index + width + x ] = p->src[index + width - x - 1];
  223. }
  224. } else {
  225. for (x = 0; x < 8; x++) {
  226. psrc16[index - x - 1] = psrc16[index + x ];
  227. psrc16[index + width + x ] = psrc16[index + width - x - 1];
  228. }
  229. }
  230. }
  231. for (y = 0; y < 8; y++) {
  232. memcpy(p->src + ( 7-y)*linesize * sample_bytes, p->src + ( y+8)*linesize * sample_bytes, linesize * sample_bytes);
  233. memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
  234. }
  235. for (y = 0; y < height + 8; y += 8) {
  236. memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
  237. for (x = 0; x < width + 8; x += 8) {
  238. int qp;
  239. if (p->qp) {
  240. qp = p->qp;
  241. } else{
  242. const int qps = 3 + is_luma;
  243. qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
  244. qp = FFMAX(1, ff_norm_qscale(qp, p->qscale_type));
  245. }
  246. for (i = 0; i < count; i++) {
  247. const int x1 = x + offset[i + count - 1][0];
  248. const int y1 = y + offset[i + count - 1][1];
  249. const int index = x1 + y1*linesize;
  250. p->dct->get_pixels(block, p->src + sample_bytes*index, sample_bytes*linesize);
  251. p->dct->fdct(block);
  252. p->requantize(block2, block, qp, p->dct->idct_permutation);
  253. p->dct->idct(block2);
  254. add_block(p->temp + index, linesize, block2);
  255. }
  256. }
  257. if (y) {
  258. if (sample_bytes == 1) {
  259. p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
  260. dst_linesize, linesize, width,
  261. FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
  262. ldither);
  263. } else {
  264. store_slice16_c((uint16_t*)(dst + (y - 8) * dst_linesize), p->temp + 8 + y*linesize,
  265. dst_linesize/2, linesize, width,
  266. FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
  267. ldither, depth);
  268. }
  269. }
  270. }
  271. }
  272. static int query_formats(AVFilterContext *ctx)
  273. {
  274. static const enum AVPixelFormat pix_fmts[] = {
  275. AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
  276. AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
  277. AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
  278. AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
  279. AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
  280. AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
  281. AV_PIX_FMT_YUV420P10,
  282. AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9,
  283. AV_PIX_FMT_YUV420P9,
  284. AV_PIX_FMT_GRAY8,
  285. AV_PIX_FMT_GBRP,
  286. AV_PIX_FMT_GBRP9,
  287. AV_PIX_FMT_GBRP10,
  288. AV_PIX_FMT_NONE
  289. };
  290. AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
  291. if (!fmts_list)
  292. return AVERROR(ENOMEM);
  293. return ff_set_common_formats(ctx, fmts_list);
  294. }
  295. static int config_input(AVFilterLink *inlink)
  296. {
  297. SPPContext *s = inlink->dst->priv;
  298. const int h = FFALIGN(inlink->h + 16, 16);
  299. const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
  300. const int bps = desc->comp[0].depth;
  301. av_opt_set_int(s->dct, "bits_per_sample", bps, 0);
  302. avcodec_dct_init(s->dct);
  303. if (ARCH_X86)
  304. ff_spp_init_x86(s);
  305. s->hsub = desc->log2_chroma_w;
  306. s->vsub = desc->log2_chroma_h;
  307. s->temp_linesize = FFALIGN(inlink->w + 16, 16);
  308. s->temp = av_malloc_array(s->temp_linesize, h * sizeof(*s->temp));
  309. s->src = av_malloc_array(s->temp_linesize, h * sizeof(*s->src) * 2);
  310. if (!s->temp || !s->src)
  311. return AVERROR(ENOMEM);
  312. return 0;
  313. }
  314. static int filter_frame(AVFilterLink *inlink, AVFrame *in)
  315. {
  316. AVFilterContext *ctx = inlink->dst;
  317. SPPContext *s = ctx->priv;
  318. AVFilterLink *outlink = ctx->outputs[0];
  319. AVFrame *out = in;
  320. int qp_stride = 0;
  321. const int8_t *qp_table = NULL;
  322. const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
  323. const int depth = desc->comp[0].depth;
  324. /* if we are not in a constant user quantizer mode and we don't want to use
  325. * the quantizers from the B-frames (B-frames often have a higher QP), we
  326. * need to save the qp table from the last non B-frame; this is what the
  327. * following code block does */
  328. if (!s->qp) {
  329. qp_table = av_frame_get_qp_table(in, &qp_stride, &s->qscale_type);
  330. if (qp_table && !s->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
  331. int w, h;
  332. /* if the qp stride is not set, it means the QP are only defined on
  333. * a line basis */
  334. if (!qp_stride) {
  335. w = AV_CEIL_RSHIFT(inlink->w, 4);
  336. h = 1;
  337. } else {
  338. w = qp_stride;
  339. h = AV_CEIL_RSHIFT(inlink->h, 4);
  340. }
  341. if (w * h > s->non_b_qp_alloc_size) {
  342. int ret = av_reallocp_array(&s->non_b_qp_table, w, h);
  343. if (ret < 0) {
  344. s->non_b_qp_alloc_size = 0;
  345. return ret;
  346. }
  347. s->non_b_qp_alloc_size = w * h;
  348. }
  349. av_assert0(w * h <= s->non_b_qp_alloc_size);
  350. memcpy(s->non_b_qp_table, qp_table, w * h);
  351. }
  352. }
  353. if (s->log2_count && !ctx->is_disabled) {
  354. if (!s->use_bframe_qp && s->non_b_qp_table)
  355. qp_table = s->non_b_qp_table;
  356. if (qp_table || s->qp) {
  357. const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
  358. const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
  359. /* get a new frame if in-place is not possible or if the dimensions
  360. * are not multiple of 8 */
  361. if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
  362. const int aligned_w = FFALIGN(inlink->w, 8);
  363. const int aligned_h = FFALIGN(inlink->h, 8);
  364. out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
  365. if (!out) {
  366. av_frame_free(&in);
  367. return AVERROR(ENOMEM);
  368. }
  369. av_frame_copy_props(out, in);
  370. out->width = in->width;
  371. out->height = in->height;
  372. }
  373. filter(s, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, depth);
  374. if (out->data[2]) {
  375. filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
  376. filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
  377. }
  378. emms_c();
  379. }
  380. }
  381. if (in != out) {
  382. if (in->data[3])
  383. av_image_copy_plane(out->data[3], out->linesize[3],
  384. in ->data[3], in ->linesize[3],
  385. inlink->w, inlink->h);
  386. av_frame_free(&in);
  387. }
  388. return ff_filter_frame(outlink, out);
  389. }
  390. static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
  391. char *res, int res_len, int flags)
  392. {
  393. SPPContext *s = ctx->priv;
  394. if (!strcmp(cmd, "level") || !strcmp(cmd, "quality")) {
  395. if (!strcmp(args, "max"))
  396. s->log2_count = MAX_LEVEL;
  397. else
  398. s->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
  399. return 0;
  400. }
  401. return AVERROR(ENOSYS);
  402. }
  403. static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
  404. {
  405. SPPContext *s = ctx->priv;
  406. int ret;
  407. s->avctx = avcodec_alloc_context3(NULL);
  408. s->dct = avcodec_dct_alloc();
  409. if (!s->avctx || !s->dct)
  410. return AVERROR(ENOMEM);
  411. if (opts) {
  412. AVDictionaryEntry *e = NULL;
  413. while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) {
  414. if ((ret = av_opt_set(s->dct, e->key, e->value, 0)) < 0)
  415. return ret;
  416. }
  417. av_dict_free(opts);
  418. }
  419. s->store_slice = store_slice_c;
  420. switch (s->mode) {
  421. case MODE_HARD: s->requantize = hardthresh_c; break;
  422. case MODE_SOFT: s->requantize = softthresh_c; break;
  423. }
  424. return 0;
  425. }
  426. static av_cold void uninit(AVFilterContext *ctx)
  427. {
  428. SPPContext *s = ctx->priv;
  429. av_freep(&s->temp);
  430. av_freep(&s->src);
  431. if (s->avctx) {
  432. avcodec_close(s->avctx);
  433. av_freep(&s->avctx);
  434. }
  435. av_freep(&s->dct);
  436. av_freep(&s->non_b_qp_table);
  437. }
  438. static const AVFilterPad spp_inputs[] = {
  439. {
  440. .name = "default",
  441. .type = AVMEDIA_TYPE_VIDEO,
  442. .config_props = config_input,
  443. .filter_frame = filter_frame,
  444. },
  445. { NULL }
  446. };
  447. static const AVFilterPad spp_outputs[] = {
  448. {
  449. .name = "default",
  450. .type = AVMEDIA_TYPE_VIDEO,
  451. },
  452. { NULL }
  453. };
  454. AVFilter ff_vf_spp = {
  455. .name = "spp",
  456. .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."),
  457. .priv_size = sizeof(SPPContext),
  458. .init_dict = init_dict,
  459. .uninit = uninit,
  460. .query_formats = query_formats,
  461. .inputs = spp_inputs,
  462. .outputs = spp_outputs,
  463. .process_command = process_command,
  464. .priv_class = &spp_class,
  465. .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
  466. };