1 /*
2 * Copyright (c) 2005 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2014 Arwa Arif <arwaarif1994@gmail.com>
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 * Postprocessing filter - 7
25 *
26 * Originally written by Michael Niedermayer for the MPlayer
27 * project, and ported by Arwa Arif for FFmpeg.
28 */
29
30 #include "libavutil/imgutils.h"
31 #include "libavutil/mem_internal.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/pixdesc.h"
34 #include "internal.h"
35 #include "qp_table.h"
36 #include "vf_pp7.h"
37
38 enum mode {
39 MODE_HARD,
40 MODE_SOFT,
41 MODE_MEDIUM
42 };
43
44 #define OFFSET(x) offsetof(PP7Context, x)
45 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
46 static const AVOption pp7_options[] = {
47 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 64, FLAGS },
48 { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_MEDIUM}, 0, 2, FLAGS, "mode" },
49 { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
50 { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
51 { "medium", "medium thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_MEDIUM}, INT_MIN, INT_MAX, FLAGS, "mode" },
52 { NULL }
53 };
54
55 AVFILTER_DEFINE_CLASS(pp7);
56
57 DECLARE_ALIGNED(8, static const uint8_t, dither)[8][8] = {
58 { 0, 48, 12, 60, 3, 51, 15, 63, },
59 { 32, 16, 44, 28, 35, 19, 47, 31, },
60 { 8, 56, 4, 52, 11, 59, 7, 55, },
61 { 40, 24, 36, 20, 43, 27, 39, 23, },
62 { 2, 50, 14, 62, 1, 49, 13, 61, },
63 { 34, 18, 46, 30, 33, 17, 45, 29, },
64 { 10, 58, 6, 54, 9, 57, 5, 53, },
65 { 42, 26, 38, 22, 41, 25, 37, 21, },
66 };
67
68 #define N0 4
69 #define N1 5
70 #define N2 10
71 #define SN0 2
72 #define SN1 2.2360679775
73 #define SN2 3.16227766017
74 #define N (1 << 16)
75
76 static const int factor[16] = {
77 N / (N0 * N0), N / (N0 * N1), N / (N0 * N0), N / (N0 * N2),
78 N / (N1 * N0), N / (N1 * N1), N / (N1 * N0), N / (N1 * N2),
79 N / (N0 * N0), N / (N0 * N1), N / (N0 * N0), N / (N0 * N2),
80 N / (N2 * N0), N / (N2 * N1), N / (N2 * N0), N / (N2 * N2),
81 };
82
init_thres2(PP7Context * p)83 static void init_thres2(PP7Context *p)
84 {
85 int qp, i;
86 int bias = 0; //FIXME
87
88 for (qp = 0; qp < 99; qp++) {
89 for (i = 0; i < 16; i++) {
90 p->thres2[qp][i] = ((i&1) ? SN2 : SN0) * ((i&4) ? SN2 : SN0) * FFMAX(1, qp) * (1<<2) - 1 - bias;
91 }
92 }
93 }
94
dctA_c(int16_t * dst,uint8_t * src,int stride)95 static inline void dctA_c(int16_t *dst, uint8_t *src, int stride)
96 {
97 int i;
98
99 for (i = 0; i < 4; i++) {
100 int s0 = src[0 * stride] + src[6 * stride];
101 int s1 = src[1 * stride] + src[5 * stride];
102 int s2 = src[2 * stride] + src[4 * stride];
103 int s3 = src[3 * stride];
104 int s = s3 + s3;
105 s3 = s - s0;
106 s0 = s + s0;
107 s = s2 + s1;
108 s2 = s2 - s1;
109 dst[0] = s0 + s;
110 dst[2] = s0 - s;
111 dst[1] = 2 * s3 + s2;
112 dst[3] = s3 - 2 * s2;
113 src++;
114 dst += 4;
115 }
116 }
117
dctB_c(int16_t * dst,int16_t * src)118 static void dctB_c(int16_t *dst, int16_t *src)
119 {
120 int i;
121
122 for (i = 0; i < 4; i++) {
123 int s0 = src[0 * 4] + src[6 * 4];
124 int s1 = src[1 * 4] + src[5 * 4];
125 int s2 = src[2 * 4] + src[4 * 4];
126 int s3 = src[3 * 4];
127 int s = s3 + s3;
128 s3 = s - s0;
129 s0 = s + s0;
130 s = s2 + s1;
131 s2 = s2 - s1;
132 dst[0 * 4] = s0 + s;
133 dst[2 * 4] = s0 - s;
134 dst[1 * 4] = 2 * s3 + s2;
135 dst[3 * 4] = s3 - 2 * s2;
136 src++;
137 dst++;
138 }
139 }
140
hardthresh_c(PP7Context * p,int16_t * src,int qp)141 static int hardthresh_c(PP7Context *p, int16_t *src, int qp)
142 {
143 int i;
144 int a;
145
146 a = src[0] * factor[0];
147 for (i = 1; i < 16; i++) {
148 unsigned int threshold1 = p->thres2[qp][i];
149 unsigned int threshold2 = threshold1 << 1;
150 int level = src[i];
151 if (((unsigned)(level + threshold1)) > threshold2)
152 a += level * factor[i];
153 }
154 return (a + (1 << 11)) >> 12;
155 }
156
mediumthresh_c(PP7Context * p,int16_t * src,int qp)157 static int mediumthresh_c(PP7Context *p, int16_t *src, int qp)
158 {
159 int i;
160 int a;
161
162 a = src[0] * factor[0];
163 for (i = 1; i < 16; i++) {
164 unsigned int threshold1 = p->thres2[qp][i];
165 unsigned int threshold2 = threshold1 << 1;
166 int level = src[i];
167 if (((unsigned)(level + threshold1)) > threshold2) {
168 if (((unsigned)(level + 2 * threshold1)) > 2 * threshold2)
169 a += level * factor[i];
170 else {
171 if (level > 0)
172 a += 2 * (level - (int)threshold1) * factor[i];
173 else
174 a += 2 * (level + (int)threshold1) * factor[i];
175 }
176 }
177 }
178 return (a + (1 << 11)) >> 12;
179 }
180
softthresh_c(PP7Context * p,int16_t * src,int qp)181 static int softthresh_c(PP7Context *p, int16_t *src, int qp)
182 {
183 int i;
184 int a;
185
186 a = src[0] * factor[0];
187 for (i = 1; i < 16; i++) {
188 unsigned int threshold1 = p->thres2[qp][i];
189 unsigned int threshold2 = threshold1 << 1;
190 int level = src[i];
191 if (((unsigned)(level + threshold1)) > threshold2) {
192 if (level > 0)
193 a += (level - (int)threshold1) * factor[i];
194 else
195 a += (level + (int)threshold1) * factor[i];
196 }
197 }
198 return (a + (1 << 11)) >> 12;
199 }
200
filter(PP7Context * p,uint8_t * dst,uint8_t * src,int dst_stride,int src_stride,int width,int height,uint8_t * qp_store,int qp_stride,int is_luma)201 static void filter(PP7Context *p, uint8_t *dst, uint8_t *src,
202 int dst_stride, int src_stride,
203 int width, int height,
204 uint8_t *qp_store, int qp_stride, int is_luma)
205 {
206 int x, y;
207 const int stride = is_luma ? p->temp_stride : ((width + 16 + 15) & (~15));
208 uint8_t *p_src = p->src + 8 * stride;
209 int16_t *block = (int16_t *)p->src;
210 int16_t *temp = (int16_t *)(p->src + 32);
211
212 if (!src || !dst) return;
213 for (y = 0; y < height; y++) {
214 int index = 8 + 8 * stride + y * stride;
215 memcpy(p_src + index, src + y * src_stride, width);
216 for (x = 0; x < 8; x++) {
217 p_src[index - x - 1]= p_src[index + x ];
218 p_src[index + width + x ]= p_src[index + width - x - 1];
219 }
220 }
221 for (y = 0; y < 8; y++) {
222 memcpy(p_src + ( 7 - y ) * stride, p_src + ( y + 8 ) * stride, stride);
223 memcpy(p_src + (height + 8 + y) * stride, p_src + (height - y + 7) * stride, stride);
224 }
225 //FIXME (try edge emu)
226
227 for (y = 0; y < height; y++) {
228 for (x = -8; x < 0; x += 4) {
229 const int index = x + y * stride + (8 - 3) * (1 + stride) + 8; //FIXME silly offset
230 uint8_t *src = p_src + index;
231 int16_t *tp = temp + 4 * x;
232
233 dctA_c(tp + 4 * 8, src, stride);
234 }
235 for (x = 0; x < width; ) {
236 const int qps = 3 + is_luma;
237 int qp;
238 int end = FFMIN(x + 8, width);
239
240 if (p->qp)
241 qp = p->qp;
242 else {
243 qp = qp_store[ (FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
244 qp = ff_norm_qscale(qp, p->qscale_type);
245 }
246 for (; x < end; x++) {
247 const int index = x + y * stride + (8 - 3) * (1 + stride) + 8; //FIXME silly offset
248 uint8_t *src = p_src + index;
249 int16_t *tp = temp + 4 * x;
250 int v;
251
252 if ((x & 3) == 0)
253 dctA_c(tp + 4 * 8, src, stride);
254
255 p->dctB(block, tp);
256
257 v = p->requantize(p, block, qp);
258 v = (v + dither[y & 7][x & 7]) >> 6;
259 if ((unsigned)v > 255)
260 v = (-v) >> 31;
261 dst[x + y * dst_stride] = v;
262 }
263 }
264 }
265 }
266
267 static const enum AVPixelFormat pix_fmts[] = {
268 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
269 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
270 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
271 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
272 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
273 AV_PIX_FMT_GBRP,
274 AV_PIX_FMT_GRAY8, AV_PIX_FMT_NONE
275 };
276
config_input(AVFilterLink * inlink)277 static int config_input(AVFilterLink *inlink)
278 {
279 AVFilterContext *ctx = inlink->dst;
280 PP7Context *pp7 = ctx->priv;
281 const int h = FFALIGN(inlink->h + 16, 16);
282 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
283
284 pp7->hsub = desc->log2_chroma_w;
285 pp7->vsub = desc->log2_chroma_h;
286
287 pp7->temp_stride = FFALIGN(inlink->w + 16, 16);
288 pp7->src = av_malloc_array(pp7->temp_stride, (h + 8) * sizeof(uint8_t));
289
290 if (!pp7->src)
291 return AVERROR(ENOMEM);
292
293 init_thres2(pp7);
294
295 switch (pp7->mode) {
296 case 0: pp7->requantize = hardthresh_c; break;
297 case 1: pp7->requantize = softthresh_c; break;
298 default:
299 case 2: pp7->requantize = mediumthresh_c; break;
300 }
301
302 pp7->dctB = dctB_c;
303
304 #if ARCH_X86
305 ff_pp7_init_x86(pp7);
306 #endif
307
308 return 0;
309 }
310
filter_frame(AVFilterLink * inlink,AVFrame * in)311 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
312 {
313 AVFilterContext *ctx = inlink->dst;
314 PP7Context *pp7 = ctx->priv;
315 AVFilterLink *outlink = ctx->outputs[0];
316 AVFrame *out = in;
317
318 int qp_stride = 0;
319 int8_t *qp_table = NULL;
320
321 if (!pp7->qp) {
322 int ret = ff_qp_table_extract(in, &qp_table, &qp_stride, NULL, &pp7->qscale_type);
323 if (ret < 0) {
324 av_frame_free(&in);
325 return ret;
326 }
327 }
328
329 if (!ctx->is_disabled) {
330 const int cw = AV_CEIL_RSHIFT(inlink->w, pp7->hsub);
331 const int ch = AV_CEIL_RSHIFT(inlink->h, pp7->vsub);
332
333 /* get a new frame if in-place is not possible or if the dimensions
334 * are not multiple of 8 */
335 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
336 const int aligned_w = FFALIGN(inlink->w, 8);
337 const int aligned_h = FFALIGN(inlink->h, 8);
338
339 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
340 if (!out) {
341 av_frame_free(&in);
342 av_freep(&qp_table);
343 return AVERROR(ENOMEM);
344 }
345 av_frame_copy_props(out, in);
346 out->width = in->width;
347 out->height = in->height;
348 }
349
350 if (qp_table || pp7->qp) {
351
352 filter(pp7, out->data[0], in->data[0], out->linesize[0], in->linesize[0],
353 inlink->w, inlink->h, qp_table, qp_stride, 1);
354 filter(pp7, out->data[1], in->data[1], out->linesize[1], in->linesize[1],
355 cw, ch, qp_table, qp_stride, 0);
356 filter(pp7, out->data[2], in->data[2], out->linesize[2], in->linesize[2],
357 cw, ch, qp_table, qp_stride, 0);
358 emms_c();
359 }
360 }
361
362 if (in != out) {
363 if (in->data[3])
364 av_image_copy_plane(out->data[3], out->linesize[3],
365 in ->data[3], in ->linesize[3],
366 inlink->w, inlink->h);
367 av_frame_free(&in);
368 }
369 av_freep(&qp_table);
370 return ff_filter_frame(outlink, out);
371 }
372
uninit(AVFilterContext * ctx)373 static av_cold void uninit(AVFilterContext *ctx)
374 {
375 PP7Context *pp7 = ctx->priv;
376 av_freep(&pp7->src);
377 }
378
379 static const AVFilterPad pp7_inputs[] = {
380 {
381 .name = "default",
382 .type = AVMEDIA_TYPE_VIDEO,
383 .config_props = config_input,
384 .filter_frame = filter_frame,
385 },
386 };
387
388 static const AVFilterPad pp7_outputs[] = {
389 {
390 .name = "default",
391 .type = AVMEDIA_TYPE_VIDEO,
392 },
393 };
394
395 const AVFilter ff_vf_pp7 = {
396 .name = "pp7",
397 .description = NULL_IF_CONFIG_SMALL("Apply Postprocessing 7 filter."),
398 .priv_size = sizeof(PP7Context),
399 .uninit = uninit,
400 FILTER_INPUTS(pp7_inputs),
401 FILTER_OUTPUTS(pp7_outputs),
402 FILTER_PIXFMTS_ARRAY(pix_fmts),
403 .priv_class = &pp7_class,
404 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
405 };
406