1 |
|
|
/* |
2 |
|
|
* Copyright (c) 2017 Paul B Mahol |
3 |
|
|
* |
4 |
|
|
* This file is part of FFmpeg. |
5 |
|
|
* |
6 |
|
|
* FFmpeg is free software; you can redistribute it and/or |
7 |
|
|
* modify it under the terms of the GNU Lesser General Public |
8 |
|
|
* License as published by the Free Software Foundation; either |
9 |
|
|
* version 2.1 of the License, or (at your option) any later version. |
10 |
|
|
* |
11 |
|
|
* FFmpeg is distributed in the hope that it will be useful, |
12 |
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 |
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 |
|
|
* Lesser General Public License for more details. |
15 |
|
|
* |
16 |
|
|
* You should have received a copy of the GNU Lesser General Public |
17 |
|
|
* License along with FFmpeg; if not, write to the Free Software |
18 |
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 |
|
|
*/ |
20 |
|
|
|
21 |
|
|
#include <float.h> |
22 |
|
|
|
23 |
|
|
#include "libavutil/imgutils.h" |
24 |
|
|
#include "libavutil/opt.h" |
25 |
|
|
#include "libavutil/pixdesc.h" |
26 |
|
|
#include "libavcodec/avfft.h" |
27 |
|
|
|
28 |
|
|
#include "avfilter.h" |
29 |
|
|
#include "formats.h" |
30 |
|
|
#include "framesync.h" |
31 |
|
|
#include "internal.h" |
32 |
|
|
#include "video.h" |
33 |
|
|
|
34 |
|
|
#define MAX_THREADS 16 |
35 |
|
|
|
36 |
|
|
typedef struct ConvolveContext { |
37 |
|
|
const AVClass *class; |
38 |
|
|
FFFrameSync fs; |
39 |
|
|
|
40 |
|
|
FFTContext *fft[4][MAX_THREADS]; |
41 |
|
|
FFTContext *ifft[4][MAX_THREADS]; |
42 |
|
|
|
43 |
|
|
int fft_bits[4]; |
44 |
|
|
int fft_len[4]; |
45 |
|
|
int planewidth[4]; |
46 |
|
|
int planeheight[4]; |
47 |
|
|
|
48 |
|
|
FFTComplex *fft_hdata[4]; |
49 |
|
|
FFTComplex *fft_vdata[4]; |
50 |
|
|
FFTComplex *fft_hdata_impulse[4]; |
51 |
|
|
FFTComplex *fft_vdata_impulse[4]; |
52 |
|
|
|
53 |
|
|
int depth; |
54 |
|
|
int planes; |
55 |
|
|
int impulse; |
56 |
|
|
float noise; |
57 |
|
|
int nb_planes; |
58 |
|
|
int got_impulse[4]; |
59 |
|
|
|
60 |
|
|
int (*filter)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs); |
61 |
|
|
} ConvolveContext; |
62 |
|
|
|
63 |
|
|
#define OFFSET(x) offsetof(ConvolveContext, x) |
64 |
|
|
#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM |
65 |
|
|
|
66 |
|
|
static const AVOption convolve_options[] = { |
67 |
|
|
{ "planes", "set planes to convolve", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=7}, 0, 15, FLAGS }, |
68 |
|
|
{ "impulse", "when to process impulses", OFFSET(impulse), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "impulse" }, |
69 |
|
|
{ "first", "process only first impulse, ignore rest", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "impulse" }, |
70 |
|
|
{ "all", "process all impulses", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "impulse" }, |
71 |
|
|
{ "noise", "set noise", OFFSET(noise), AV_OPT_TYPE_FLOAT, {.dbl=0.0000001}, 0, 1, FLAGS }, |
72 |
|
|
{ NULL }, |
73 |
|
|
}; |
74 |
|
|
|
75 |
|
|
static int query_formats(AVFilterContext *ctx) |
76 |
|
|
{ |
77 |
|
|
static const enum AVPixelFormat pixel_fmts_fftfilt[] = { |
78 |
|
|
AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P, |
79 |
|
|
AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, |
80 |
|
|
AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P, |
81 |
|
|
AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P, |
82 |
|
|
AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P, |
83 |
|
|
AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9, |
84 |
|
|
AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, |
85 |
|
|
AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12, |
86 |
|
|
AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14, |
87 |
|
|
AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16, |
88 |
|
|
AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9, |
89 |
|
|
AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10, |
90 |
|
|
AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16, |
91 |
|
|
AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, |
92 |
|
|
AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16, |
93 |
|
|
AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16, |
94 |
|
|
AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16, |
95 |
|
|
AV_PIX_FMT_NONE |
96 |
|
|
}; |
97 |
|
|
|
98 |
|
|
AVFilterFormats *fmts_list = ff_make_format_list(pixel_fmts_fftfilt); |
99 |
|
|
if (!fmts_list) |
100 |
|
|
return AVERROR(ENOMEM); |
101 |
|
|
return ff_set_common_formats(ctx, fmts_list); |
102 |
|
|
} |
103 |
|
|
|
104 |
|
|
static int config_input_main(AVFilterLink *inlink) |
105 |
|
|
{ |
106 |
|
|
ConvolveContext *s = inlink->dst->priv; |
107 |
|
|
const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); |
108 |
|
|
int fft_bits, i; |
109 |
|
|
|
110 |
|
|
s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w); |
111 |
|
|
s->planewidth[0] = s->planewidth[3] = inlink->w; |
112 |
|
|
s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h); |
113 |
|
|
s->planeheight[0] = s->planeheight[3] = inlink->h; |
114 |
|
|
|
115 |
|
|
s->nb_planes = desc->nb_components; |
116 |
|
|
s->depth = desc->comp[0].depth; |
117 |
|
|
|
118 |
|
|
for (i = 0; i < s->nb_planes; i++) { |
119 |
|
|
int w = s->planewidth[i]; |
120 |
|
|
int h = s->planeheight[i]; |
121 |
|
|
int n = FFMAX(w, h); |
122 |
|
|
|
123 |
|
|
for (fft_bits = 1; 1 << fft_bits < n; fft_bits++); |
124 |
|
|
|
125 |
|
|
s->fft_bits[i] = fft_bits; |
126 |
|
|
s->fft_len[i] = 1 << s->fft_bits[i]; |
127 |
|
|
|
128 |
|
|
if (!(s->fft_hdata[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex)))) |
129 |
|
|
return AVERROR(ENOMEM); |
130 |
|
|
|
131 |
|
|
if (!(s->fft_vdata[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex)))) |
132 |
|
|
return AVERROR(ENOMEM); |
133 |
|
|
|
134 |
|
|
if (!(s->fft_hdata_impulse[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex)))) |
135 |
|
|
return AVERROR(ENOMEM); |
136 |
|
|
|
137 |
|
|
if (!(s->fft_vdata_impulse[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex)))) |
138 |
|
|
return AVERROR(ENOMEM); |
139 |
|
|
} |
140 |
|
|
|
141 |
|
|
return 0; |
142 |
|
|
} |
143 |
|
|
|
144 |
|
|
static int config_input_impulse(AVFilterLink *inlink) |
145 |
|
|
{ |
146 |
|
|
AVFilterContext *ctx = inlink->dst; |
147 |
|
|
|
148 |
|
|
if (ctx->inputs[0]->w != ctx->inputs[1]->w || |
149 |
|
|
ctx->inputs[0]->h != ctx->inputs[1]->h) { |
150 |
|
|
av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n"); |
151 |
|
|
return AVERROR(EINVAL); |
152 |
|
|
} |
153 |
|
|
if (ctx->inputs[0]->format != ctx->inputs[1]->format) { |
154 |
|
|
av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n"); |
155 |
|
|
return AVERROR(EINVAL); |
156 |
|
|
} |
157 |
|
|
|
158 |
|
|
return 0; |
159 |
|
|
} |
160 |
|
|
|
161 |
|
|
typedef struct ThreadData { |
162 |
|
|
FFTComplex *hdata, *vdata; |
163 |
|
|
int plane, n; |
164 |
|
|
} ThreadData; |
165 |
|
|
|
166 |
|
|
static int fft_horizontal(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
167 |
|
|
{ |
168 |
|
|
ConvolveContext *s = ctx->priv; |
169 |
|
|
ThreadData *td = arg; |
170 |
|
|
FFTComplex *hdata = td->hdata; |
171 |
|
|
const int plane = td->plane; |
172 |
|
|
const int n = td->n; |
173 |
|
|
int start = (n * jobnr) / nb_jobs; |
174 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
175 |
|
|
int y; |
176 |
|
|
|
177 |
|
|
for (y = start; y < end; y++) { |
178 |
|
|
av_fft_permute(s->fft[plane][jobnr], hdata + y * n); |
179 |
|
|
av_fft_calc(s->fft[plane][jobnr], hdata + y * n); |
180 |
|
|
} |
181 |
|
|
|
182 |
|
|
return 0; |
183 |
|
|
} |
184 |
|
|
|
185 |
|
|
static void get_input(ConvolveContext *s, FFTComplex *fft_hdata, |
186 |
|
|
AVFrame *in, int w, int h, int n, int plane, float scale) |
187 |
|
|
{ |
188 |
|
|
const int iw = (n - w) / 2, ih = (n - h) / 2; |
189 |
|
|
int y, x; |
190 |
|
|
|
191 |
|
|
if (s->depth == 8) { |
192 |
|
|
for (y = 0; y < h; y++) { |
193 |
|
|
const uint8_t *src = in->data[plane] + in->linesize[plane] * y; |
194 |
|
|
|
195 |
|
|
for (x = 0; x < w; x++) { |
196 |
|
|
fft_hdata[(y + ih) * n + iw + x].re = src[x] * scale; |
197 |
|
|
fft_hdata[(y + ih) * n + iw + x].im = 0; |
198 |
|
|
} |
199 |
|
|
|
200 |
|
|
for (x = 0; x < iw; x++) { |
201 |
|
|
fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + iw].re; |
202 |
|
|
fft_hdata[(y + ih) * n + x].im = 0; |
203 |
|
|
} |
204 |
|
|
|
205 |
|
|
for (x = n - iw; x < n; x++) { |
206 |
|
|
fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + n - iw - 1].re; |
207 |
|
|
fft_hdata[(y + ih) * n + x].im = 0; |
208 |
|
|
} |
209 |
|
|
} |
210 |
|
|
|
211 |
|
|
for (y = 0; y < ih; y++) { |
212 |
|
|
for (x = 0; x < n; x++) { |
213 |
|
|
fft_hdata[y * n + x].re = fft_hdata[ih * n + x].re; |
214 |
|
|
fft_hdata[y * n + x].im = 0; |
215 |
|
|
} |
216 |
|
|
} |
217 |
|
|
|
218 |
|
|
for (y = n - ih; y < n; y++) { |
219 |
|
|
for (x = 0; x < n; x++) { |
220 |
|
|
fft_hdata[y * n + x].re = fft_hdata[(n - ih - 1) * n + x].re; |
221 |
|
|
fft_hdata[y * n + x].im = 0; |
222 |
|
|
} |
223 |
|
|
} |
224 |
|
|
} else { |
225 |
|
|
for (y = 0; y < h; y++) { |
226 |
|
|
const uint16_t *src = (const uint16_t *)(in->data[plane] + in->linesize[plane] * y); |
227 |
|
|
|
228 |
|
|
for (x = 0; x < w; x++) { |
229 |
|
|
fft_hdata[(y + ih) * n + iw + x].re = src[x] * scale; |
230 |
|
|
fft_hdata[(y + ih) * n + iw + x].im = 0; |
231 |
|
|
} |
232 |
|
|
|
233 |
|
|
for (x = 0; x < iw; x++) { |
234 |
|
|
fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + iw].re; |
235 |
|
|
fft_hdata[(y + ih) * n + x].im = 0; |
236 |
|
|
} |
237 |
|
|
|
238 |
|
|
for (x = n - iw; x < n; x++) { |
239 |
|
|
fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + n - iw - 1].re; |
240 |
|
|
fft_hdata[(y + ih) * n + x].im = 0; |
241 |
|
|
} |
242 |
|
|
} |
243 |
|
|
|
244 |
|
|
for (y = 0; y < ih; y++) { |
245 |
|
|
for (x = 0; x < n; x++) { |
246 |
|
|
fft_hdata[y * n + x].re = fft_hdata[ih * n + x].re; |
247 |
|
|
fft_hdata[y * n + x].im = 0; |
248 |
|
|
} |
249 |
|
|
} |
250 |
|
|
|
251 |
|
|
for (y = n - ih; y < n; y++) { |
252 |
|
|
for (x = 0; x < n; x++) { |
253 |
|
|
fft_hdata[y * n + x].re = fft_hdata[(n - ih - 1) * n + x].re; |
254 |
|
|
fft_hdata[y * n + x].im = 0; |
255 |
|
|
} |
256 |
|
|
} |
257 |
|
|
} |
258 |
|
|
} |
259 |
|
|
|
260 |
|
|
static int fft_vertical(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
261 |
|
|
{ |
262 |
|
|
ConvolveContext *s = ctx->priv; |
263 |
|
|
ThreadData *td = arg; |
264 |
|
|
FFTComplex *hdata = td->hdata; |
265 |
|
|
FFTComplex *vdata = td->vdata; |
266 |
|
|
const int plane = td->plane; |
267 |
|
|
const int n = td->n; |
268 |
|
|
int start = (n * jobnr) / nb_jobs; |
269 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
270 |
|
|
int y, x; |
271 |
|
|
|
272 |
|
|
for (y = start; y < end; y++) { |
273 |
|
|
for (x = 0; x < n; x++) { |
274 |
|
|
vdata[y * n + x].re = hdata[x * n + y].re; |
275 |
|
|
vdata[y * n + x].im = hdata[x * n + y].im; |
276 |
|
|
} |
277 |
|
|
|
278 |
|
|
av_fft_permute(s->fft[plane][jobnr], vdata + y * n); |
279 |
|
|
av_fft_calc(s->fft[plane][jobnr], vdata + y * n); |
280 |
|
|
} |
281 |
|
|
|
282 |
|
|
return 0; |
283 |
|
|
} |
284 |
|
|
|
285 |
|
|
static int ifft_vertical(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
286 |
|
|
{ |
287 |
|
|
ConvolveContext *s = ctx->priv; |
288 |
|
|
ThreadData *td = arg; |
289 |
|
|
FFTComplex *hdata = td->hdata; |
290 |
|
|
FFTComplex *vdata = td->vdata; |
291 |
|
|
const int plane = td->plane; |
292 |
|
|
const int n = td->n; |
293 |
|
|
int start = (n * jobnr) / nb_jobs; |
294 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
295 |
|
|
int y, x; |
296 |
|
|
|
297 |
|
|
for (y = start; y < end; y++) { |
298 |
|
|
av_fft_permute(s->ifft[plane][jobnr], vdata + y * n); |
299 |
|
|
av_fft_calc(s->ifft[plane][jobnr], vdata + y * n); |
300 |
|
|
|
301 |
|
|
for (x = 0; x < n; x++) { |
302 |
|
|
hdata[x * n + y].re = vdata[y * n + x].re; |
303 |
|
|
hdata[x * n + y].im = vdata[y * n + x].im; |
304 |
|
|
} |
305 |
|
|
} |
306 |
|
|
|
307 |
|
|
return 0; |
308 |
|
|
} |
309 |
|
|
|
310 |
|
|
static int ifft_horizontal(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
311 |
|
|
{ |
312 |
|
|
ConvolveContext *s = ctx->priv; |
313 |
|
|
ThreadData *td = arg; |
314 |
|
|
FFTComplex *hdata = td->hdata; |
315 |
|
|
const int plane = td->plane; |
316 |
|
|
const int n = td->n; |
317 |
|
|
int start = (n * jobnr) / nb_jobs; |
318 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
319 |
|
|
int y; |
320 |
|
|
|
321 |
|
|
for (y = start; y < end; y++) { |
322 |
|
|
av_fft_permute(s->ifft[plane][jobnr], hdata + y * n); |
323 |
|
|
av_fft_calc(s->ifft[plane][jobnr], hdata + y * n); |
324 |
|
|
} |
325 |
|
|
|
326 |
|
|
return 0; |
327 |
|
|
} |
328 |
|
|
|
329 |
|
|
static void get_output(ConvolveContext *s, FFTComplex *input, AVFrame *out, |
330 |
|
|
int w, int h, int n, int plane, float scale) |
331 |
|
|
{ |
332 |
|
|
const int max = (1 << s->depth) - 1; |
333 |
|
|
const int hh = h / 2; |
334 |
|
|
const int hw = w / 2; |
335 |
|
|
int y, x; |
336 |
|
|
|
337 |
|
|
if (s->depth == 8) { |
338 |
|
|
for (y = 0; y < hh; y++) { |
339 |
|
|
uint8_t *dst = out->data[plane] + (y + hh) * out->linesize[plane] + hw; |
340 |
|
|
for (x = 0; x < hw; x++) |
341 |
|
|
dst[x] = av_clip_uint8(input[y * n + x].re * scale); |
342 |
|
|
} |
343 |
|
|
for (y = 0; y < hh; y++) { |
344 |
|
|
uint8_t *dst = out->data[plane] + (y + hh) * out->linesize[plane]; |
345 |
|
|
for (x = 0; x < hw; x++) |
346 |
|
|
dst[x] = av_clip_uint8(input[y * n + n - hw + x].re * scale); |
347 |
|
|
} |
348 |
|
|
for (y = 0; y < hh; y++) { |
349 |
|
|
uint8_t *dst = out->data[plane] + y * out->linesize[plane] + hw; |
350 |
|
|
for (x = 0; x < hw; x++) |
351 |
|
|
dst[x] = av_clip_uint8(input[(n - hh + y) * n + x].re * scale); |
352 |
|
|
} |
353 |
|
|
for (y = 0; y < hh; y++) { |
354 |
|
|
uint8_t *dst = out->data[plane] + y * out->linesize[plane]; |
355 |
|
|
for (x = 0; x < hw; x++) |
356 |
|
|
dst[x] = av_clip_uint8(input[(n - hh + y) * n + n - hw + x].re * scale); |
357 |
|
|
} |
358 |
|
|
} else { |
359 |
|
|
for (y = 0; y < hh; y++) { |
360 |
|
|
uint16_t *dst = (uint16_t *)(out->data[plane] + (y + hh) * out->linesize[plane] + hw * 2); |
361 |
|
|
for (x = 0; x < hw; x++) |
362 |
|
|
dst[x] = av_clip(input[y * n + x].re * scale, 0, max); |
363 |
|
|
} |
364 |
|
|
for (y = 0; y < hh; y++) { |
365 |
|
|
uint16_t *dst = (uint16_t *)(out->data[plane] + (y + hh) * out->linesize[plane]); |
366 |
|
|
for (x = 0; x < hw; x++) |
367 |
|
|
dst[x] = av_clip(input[y * n + n - hw + x].re * scale, 0, max); |
368 |
|
|
} |
369 |
|
|
for (y = 0; y < hh; y++) { |
370 |
|
|
uint16_t *dst = (uint16_t *)(out->data[plane] + y * out->linesize[plane] + hw * 2); |
371 |
|
|
for (x = 0; x < hw; x++) |
372 |
|
|
dst[x] = av_clip(input[(n - hh + y) * n + x].re * scale, 0, max); |
373 |
|
|
} |
374 |
|
|
for (y = 0; y < hh; y++) { |
375 |
|
|
uint16_t *dst = (uint16_t *)(out->data[plane] + y * out->linesize[plane]); |
376 |
|
|
for (x = 0; x < hw; x++) |
377 |
|
|
dst[x] = av_clip(input[(n - hh + y) * n + n - hw + x].re * scale, 0, max); |
378 |
|
|
} |
379 |
|
|
} |
380 |
|
|
} |
381 |
|
|
|
382 |
|
|
static int complex_multiply(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
383 |
|
|
{ |
384 |
|
|
ConvolveContext *s = ctx->priv; |
385 |
|
|
ThreadData *td = arg; |
386 |
|
|
FFTComplex *input = td->hdata; |
387 |
|
|
FFTComplex *filter = td->vdata; |
388 |
|
|
const float noise = s->noise; |
389 |
|
|
const int n = td->n; |
390 |
|
|
int start = (n * jobnr) / nb_jobs; |
391 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
392 |
|
|
int y, x; |
393 |
|
|
|
394 |
|
|
for (y = start; y < end; y++) { |
395 |
|
|
int yn = y * n; |
396 |
|
|
|
397 |
|
|
for (x = 0; x < n; x++) { |
398 |
|
|
FFTSample re, im, ire, iim; |
399 |
|
|
|
400 |
|
|
re = input[yn + x].re; |
401 |
|
|
im = input[yn + x].im; |
402 |
|
|
ire = filter[yn + x].re + noise; |
403 |
|
|
iim = filter[yn + x].im; |
404 |
|
|
|
405 |
|
|
input[yn + x].re = ire * re - iim * im; |
406 |
|
|
input[yn + x].im = iim * re + ire * im; |
407 |
|
|
} |
408 |
|
|
} |
409 |
|
|
|
410 |
|
|
return 0; |
411 |
|
|
} |
412 |
|
|
|
413 |
|
|
static int complex_divide(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
414 |
|
|
{ |
415 |
|
|
ConvolveContext *s = ctx->priv; |
416 |
|
|
ThreadData *td = arg; |
417 |
|
|
FFTComplex *input = td->hdata; |
418 |
|
|
FFTComplex *filter = td->vdata; |
419 |
|
|
const float noise = s->noise; |
420 |
|
|
const int n = td->n; |
421 |
|
|
int start = (n * jobnr) / nb_jobs; |
422 |
|
|
int end = (n * (jobnr+1)) / nb_jobs; |
423 |
|
|
int y, x; |
424 |
|
|
|
425 |
|
|
for (y = start; y < end; y++) { |
426 |
|
|
int yn = y * n; |
427 |
|
|
|
428 |
|
|
for (x = 0; x < n; x++) { |
429 |
|
|
FFTSample re, im, ire, iim, div; |
430 |
|
|
|
431 |
|
|
re = input[yn + x].re; |
432 |
|
|
im = input[yn + x].im; |
433 |
|
|
ire = filter[yn + x].re; |
434 |
|
|
iim = filter[yn + x].im; |
435 |
|
|
div = ire * ire + iim * iim + noise; |
436 |
|
|
|
437 |
|
|
input[yn + x].re = (ire * re + iim * im) / div; |
438 |
|
|
input[yn + x].im = (ire * im - iim * re) / div; |
439 |
|
|
} |
440 |
|
|
} |
441 |
|
|
|
442 |
|
|
return 0; |
443 |
|
|
} |
444 |
|
|
|
445 |
|
|
static int do_convolve(FFFrameSync *fs) |
446 |
|
|
{ |
447 |
|
|
AVFilterContext *ctx = fs->parent; |
448 |
|
|
AVFilterLink *outlink = ctx->outputs[0]; |
449 |
|
|
ConvolveContext *s = ctx->priv; |
450 |
|
|
AVFrame *mainpic = NULL, *impulsepic = NULL; |
451 |
|
|
int ret, y, x, plane; |
452 |
|
|
|
453 |
|
|
ret = ff_framesync_dualinput_get(fs, &mainpic, &impulsepic); |
454 |
|
|
if (ret < 0) |
455 |
|
|
return ret; |
456 |
|
|
if (!impulsepic) |
457 |
|
|
return ff_filter_frame(outlink, mainpic); |
458 |
|
|
|
459 |
|
|
for (plane = 0; plane < s->nb_planes; plane++) { |
460 |
|
|
FFTComplex *filter = s->fft_vdata_impulse[plane]; |
461 |
|
|
FFTComplex *input = s->fft_vdata[plane]; |
462 |
|
|
const int n = s->fft_len[plane]; |
463 |
|
|
const int w = s->planewidth[plane]; |
464 |
|
|
const int h = s->planeheight[plane]; |
465 |
|
|
float total = 0; |
466 |
|
|
ThreadData td; |
467 |
|
|
|
468 |
|
|
if (!(s->planes & (1 << plane))) { |
469 |
|
|
continue; |
470 |
|
|
} |
471 |
|
|
|
472 |
|
|
td.plane = plane, td.n = n; |
473 |
|
|
get_input(s, s->fft_hdata[plane], mainpic, w, h, n, plane, 1.f); |
474 |
|
|
|
475 |
|
|
td.hdata = s->fft_hdata[plane]; |
476 |
|
|
td.vdata = s->fft_vdata[plane]; |
477 |
|
|
|
478 |
|
|
ctx->internal->execute(ctx, fft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
479 |
|
|
ctx->internal->execute(ctx, fft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
480 |
|
|
|
481 |
|
|
if ((!s->impulse && !s->got_impulse[plane]) || s->impulse) { |
482 |
|
|
if (s->depth == 8) { |
483 |
|
|
for (y = 0; y < h; y++) { |
484 |
|
|
const uint8_t *src = (const uint8_t *)(impulsepic->data[plane] + y * impulsepic->linesize[plane]) ; |
485 |
|
|
for (x = 0; x < w; x++) { |
486 |
|
|
total += src[x]; |
487 |
|
|
} |
488 |
|
|
} |
489 |
|
|
} else { |
490 |
|
|
for (y = 0; y < h; y++) { |
491 |
|
|
const uint16_t *src = (const uint16_t *)(impulsepic->data[plane] + y * impulsepic->linesize[plane]) ; |
492 |
|
|
for (x = 0; x < w; x++) { |
493 |
|
|
total += src[x]; |
494 |
|
|
} |
495 |
|
|
} |
496 |
|
|
} |
497 |
|
|
total = FFMAX(1, total); |
498 |
|
|
|
499 |
|
|
get_input(s, s->fft_hdata_impulse[plane], impulsepic, w, h, n, plane, 1.f / total); |
500 |
|
|
|
501 |
|
|
td.hdata = s->fft_hdata_impulse[plane]; |
502 |
|
|
td.vdata = s->fft_vdata_impulse[plane]; |
503 |
|
|
|
504 |
|
|
ctx->internal->execute(ctx, fft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
505 |
|
|
ctx->internal->execute(ctx, fft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
506 |
|
|
|
507 |
|
|
s->got_impulse[plane] = 1; |
508 |
|
|
} |
509 |
|
|
|
510 |
|
|
td.hdata = input; |
511 |
|
|
td.vdata = filter; |
512 |
|
|
|
513 |
|
|
ctx->internal->execute(ctx, s->filter, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
514 |
|
|
|
515 |
|
|
td.hdata = s->fft_hdata[plane]; |
516 |
|
|
td.vdata = s->fft_vdata[plane]; |
517 |
|
|
|
518 |
|
|
ctx->internal->execute(ctx, ifft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
519 |
|
|
ctx->internal->execute(ctx, ifft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx))); |
520 |
|
|
|
521 |
|
|
get_output(s, s->fft_hdata[plane], mainpic, w, h, n, plane, 1.f / (n * n)); |
522 |
|
|
} |
523 |
|
|
|
524 |
|
|
return ff_filter_frame(outlink, mainpic); |
525 |
|
|
} |
526 |
|
|
|
527 |
|
|
static int config_output(AVFilterLink *outlink) |
528 |
|
|
{ |
529 |
|
|
AVFilterContext *ctx = outlink->src; |
530 |
|
|
ConvolveContext *s = ctx->priv; |
531 |
|
|
AVFilterLink *mainlink = ctx->inputs[0]; |
532 |
|
|
int ret, i, j; |
533 |
|
|
|
534 |
|
|
s->fs.on_event = do_convolve; |
535 |
|
|
ret = ff_framesync_init_dualinput(&s->fs, ctx); |
536 |
|
|
if (ret < 0) |
537 |
|
|
return ret; |
538 |
|
|
outlink->w = mainlink->w; |
539 |
|
|
outlink->h = mainlink->h; |
540 |
|
|
outlink->time_base = mainlink->time_base; |
541 |
|
|
outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio; |
542 |
|
|
outlink->frame_rate = mainlink->frame_rate; |
543 |
|
|
|
544 |
|
|
if ((ret = ff_framesync_configure(&s->fs)) < 0) |
545 |
|
|
return ret; |
546 |
|
|
|
547 |
|
|
for (i = 0; i < s->nb_planes; i++) { |
548 |
|
|
for (j = 0; j < MAX_THREADS; j++) { |
549 |
|
|
s->fft[i][j] = av_fft_init(s->fft_bits[i], 0); |
550 |
|
|
s->ifft[i][j] = av_fft_init(s->fft_bits[i], 1); |
551 |
|
|
if (!s->fft[i][j] || !s->ifft[i][j]) |
552 |
|
|
return AVERROR(ENOMEM); |
553 |
|
|
} |
554 |
|
|
} |
555 |
|
|
|
556 |
|
|
return 0; |
557 |
|
|
} |
558 |
|
|
|
559 |
|
|
static int activate(AVFilterContext *ctx) |
560 |
|
|
{ |
561 |
|
|
ConvolveContext *s = ctx->priv; |
562 |
|
|
return ff_framesync_activate(&s->fs); |
563 |
|
|
} |
564 |
|
|
|
565 |
|
|
static av_cold int init(AVFilterContext *ctx) |
566 |
|
|
{ |
567 |
|
|
ConvolveContext *s = ctx->priv; |
568 |
|
|
|
569 |
|
|
if (!strcmp(ctx->filter->name, "convolve")) { |
570 |
|
|
s->filter = complex_multiply; |
571 |
|
|
} else if (!strcmp(ctx->filter->name, "deconvolve")) { |
572 |
|
|
s->filter = complex_divide; |
573 |
|
|
} else { |
574 |
|
|
return AVERROR_BUG; |
575 |
|
|
} |
576 |
|
|
|
577 |
|
|
return 0; |
578 |
|
|
} |
579 |
|
|
|
580 |
|
|
static av_cold void uninit(AVFilterContext *ctx) |
581 |
|
|
{ |
582 |
|
|
ConvolveContext *s = ctx->priv; |
583 |
|
|
int i, j; |
584 |
|
|
|
585 |
|
|
for (i = 0; i < 4; i++) { |
586 |
|
|
av_freep(&s->fft_hdata[i]); |
587 |
|
|
av_freep(&s->fft_vdata[i]); |
588 |
|
|
av_freep(&s->fft_hdata_impulse[i]); |
589 |
|
|
av_freep(&s->fft_vdata_impulse[i]); |
590 |
|
|
|
591 |
|
|
for (j = 0; j < MAX_THREADS; j++) { |
592 |
|
|
av_fft_end(s->fft[i][j]); |
593 |
|
|
s->fft[i][j] = NULL; |
594 |
|
|
av_fft_end(s->ifft[i][j]); |
595 |
|
|
s->ifft[i][j] = NULL; |
596 |
|
|
} |
597 |
|
|
} |
598 |
|
|
|
599 |
|
|
ff_framesync_uninit(&s->fs); |
600 |
|
|
} |
601 |
|
|
|
602 |
|
|
static const AVFilterPad convolve_inputs[] = { |
603 |
|
|
{ |
604 |
|
|
.name = "main", |
605 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
606 |
|
|
.config_props = config_input_main, |
607 |
|
|
},{ |
608 |
|
|
.name = "impulse", |
609 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
610 |
|
|
.config_props = config_input_impulse, |
611 |
|
|
}, |
612 |
|
|
{ NULL } |
613 |
|
|
}; |
614 |
|
|
|
615 |
|
|
static const AVFilterPad convolve_outputs[] = { |
616 |
|
|
{ |
617 |
|
|
.name = "default", |
618 |
|
|
.type = AVMEDIA_TYPE_VIDEO, |
619 |
|
|
.config_props = config_output, |
620 |
|
|
}, |
621 |
|
|
{ NULL } |
622 |
|
|
}; |
623 |
|
|
|
624 |
|
|
#if CONFIG_CONVOLVE_FILTER |
625 |
|
|
|
626 |
|
|
FRAMESYNC_DEFINE_CLASS(convolve, ConvolveContext, fs); |
627 |
|
|
|
628 |
|
|
AVFilter ff_vf_convolve = { |
629 |
|
|
.name = "convolve", |
630 |
|
|
.description = NULL_IF_CONFIG_SMALL("Convolve first video stream with second video stream."), |
631 |
|
|
.preinit = convolve_framesync_preinit, |
632 |
|
|
.init = init, |
633 |
|
|
.uninit = uninit, |
634 |
|
|
.query_formats = query_formats, |
635 |
|
|
.activate = activate, |
636 |
|
|
.priv_size = sizeof(ConvolveContext), |
637 |
|
|
.priv_class = &convolve_class, |
638 |
|
|
.inputs = convolve_inputs, |
639 |
|
|
.outputs = convolve_outputs, |
640 |
|
|
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, |
641 |
|
|
}; |
642 |
|
|
|
643 |
|
|
#endif /* CONFIG_CONVOLVE_FILTER */ |
644 |
|
|
|
645 |
|
|
#if CONFIG_DECONVOLVE_FILTER |
646 |
|
|
|
647 |
|
|
static const AVOption deconvolve_options[] = { |
648 |
|
|
{ "planes", "set planes to deconvolve", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=7}, 0, 15, FLAGS }, |
649 |
|
|
{ "impulse", "when to process impulses", OFFSET(impulse), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "impulse" }, |
650 |
|
|
{ "first", "process only first impulse, ignore rest", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "impulse" }, |
651 |
|
|
{ "all", "process all impulses", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "impulse" }, |
652 |
|
|
{ "noise", "set noise", OFFSET(noise), AV_OPT_TYPE_FLOAT, {.dbl=0.0000001}, 0, 1, FLAGS }, |
653 |
|
|
{ NULL }, |
654 |
|
|
}; |
655 |
|
|
|
656 |
|
|
FRAMESYNC_DEFINE_CLASS(deconvolve, ConvolveContext, fs); |
657 |
|
|
|
658 |
|
|
AVFilter ff_vf_deconvolve = { |
659 |
|
|
.name = "deconvolve", |
660 |
|
|
.description = NULL_IF_CONFIG_SMALL("Deconvolve first video stream with second video stream."), |
661 |
|
|
.preinit = deconvolve_framesync_preinit, |
662 |
|
|
.init = init, |
663 |
|
|
.uninit = uninit, |
664 |
|
|
.query_formats = query_formats, |
665 |
|
|
.activate = activate, |
666 |
|
|
.priv_size = sizeof(ConvolveContext), |
667 |
|
|
.priv_class = &deconvolve_class, |
668 |
|
|
.inputs = convolve_inputs, |
669 |
|
|
.outputs = convolve_outputs, |
670 |
|
|
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, |
671 |
|
|
}; |
672 |
|
|
|
673 |
|
|
#endif /* CONFIG_DECONVOLVE_FILTER */ |