libavfilter/vf_convolve.c
4d41db7a
 /*
  * Copyright (c) 2017 Paul B Mahol
  *
  * This file is part of FFmpeg.
  *
  * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
  * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
  * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
8c9a91ac
 #include <float.h>
 
4d41db7a
 #include "libavutil/imgutils.h"
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
 #include "libavcodec/avfft.h"
 
 #include "avfilter.h"
 #include "formats.h"
5f5dcf44
 #include "framesync.h"
4d41db7a
 #include "internal.h"
 #include "video.h"
 
c59e49f9
 #define MAX_THREADS 16
 
4d41db7a
 typedef struct ConvolveContext {
     const AVClass *class;
     FFFrameSync fs;
 
c59e49f9
     FFTContext *fft[4][MAX_THREADS];
     FFTContext *ifft[4][MAX_THREADS];
4d41db7a
 
     int fft_bits[4];
     int fft_len[4];
     int planewidth[4];
     int planeheight[4];
 
     FFTComplex *fft_hdata[4];
     FFTComplex *fft_vdata[4];
     FFTComplex *fft_hdata_impulse[4];
     FFTComplex *fft_vdata_impulse[4];
 
     int depth;
     int planes;
     int impulse;
8c9a91ac
     float noise;
4d41db7a
     int nb_planes;
     int got_impulse[4];
8c9a91ac
 
     int (*filter)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
4d41db7a
 } ConvolveContext;
 
 #define OFFSET(x) offsetof(ConvolveContext, x)
 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
 
 static const AVOption convolve_options[] = {
     { "planes",  "set planes to convolve",                  OFFSET(planes),   AV_OPT_TYPE_INT,   {.i64=7}, 0, 15, FLAGS },
     { "impulse", "when to process impulses",                OFFSET(impulse),  AV_OPT_TYPE_INT,   {.i64=1}, 0,  1, FLAGS, "impulse" },
     {   "first", "process only first impulse, ignore rest", 0,                AV_OPT_TYPE_CONST, {.i64=0}, 0,  0, FLAGS, "impulse" },
     {   "all",   "process all impulses",                    0,                AV_OPT_TYPE_CONST, {.i64=1}, 0,  0, FLAGS, "impulse" },
8c9a91ac
     { "noise",   "set noise",                               OFFSET(noise),    AV_OPT_TYPE_FLOAT, {.dbl=0.0000001}, 0,  1, FLAGS },
4d41db7a
     { NULL },
 };
 
 static int query_formats(AVFilterContext *ctx)
 {
     static const enum AVPixelFormat pixel_fmts_fftfilt[] = {
         AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
         AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
         AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
         AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
         AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
         AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
         AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
         AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
         AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
         AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
         AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
         AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
         AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
         AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
         AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
         AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
         AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY16,
         AV_PIX_FMT_NONE
     };
 
     AVFilterFormats *fmts_list = ff_make_format_list(pixel_fmts_fftfilt);
     if (!fmts_list)
         return AVERROR(ENOMEM);
     return ff_set_common_formats(ctx, fmts_list);
 }
 
 static int config_input_main(AVFilterLink *inlink)
 {
     ConvolveContext *s = inlink->dst->priv;
     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
     int fft_bits, i;
 
     s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
     s->planewidth[0] = s->planewidth[3] = inlink->w;
     s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
     s->planeheight[0] = s->planeheight[3] = inlink->h;
 
     s->nb_planes = desc->nb_components;
     s->depth = desc->comp[0].depth;
 
     for (i = 0; i < s->nb_planes; i++) {
         int w = s->planewidth[i];
         int h = s->planeheight[i];
b5958ff8
         int n = FFMAX(w, h);
 
4d41db7a
         for (fft_bits = 1; 1 << fft_bits < n; fft_bits++);
 
5533cbfc
         s->fft_bits[i] = fft_bits;
4d41db7a
         s->fft_len[i] = 1 << s->fft_bits[i];
 
87f148d5
         if (!(s->fft_hdata[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex))))
4d41db7a
             return AVERROR(ENOMEM);
 
87f148d5
         if (!(s->fft_vdata[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex))))
4d41db7a
             return AVERROR(ENOMEM);
 
87f148d5
         if (!(s->fft_hdata_impulse[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex))))
4d41db7a
             return AVERROR(ENOMEM);
 
87f148d5
         if (!(s->fft_vdata_impulse[i] = av_calloc(s->fft_len[i], s->fft_len[i] * sizeof(FFTComplex))))
4d41db7a
             return AVERROR(ENOMEM);
     }
 
     return 0;
 }
 
 static int config_input_impulse(AVFilterLink *inlink)
 {
     AVFilterContext *ctx  = inlink->dst;
 
     if (ctx->inputs[0]->w != ctx->inputs[1]->w ||
         ctx->inputs[0]->h != ctx->inputs[1]->h) {
         av_log(ctx, AV_LOG_ERROR, "Width and height of input videos must be same.\n");
         return AVERROR(EINVAL);
     }
     if (ctx->inputs[0]->format != ctx->inputs[1]->format) {
         av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel format.\n");
         return AVERROR(EINVAL);
     }
 
     return 0;
 }
 
c59e49f9
 typedef struct ThreadData {
     FFTComplex *hdata, *vdata;
     int plane, n;
 } ThreadData;
 
 static int fft_horizontal(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
f9f1b2a9
 {
c59e49f9
     ConvolveContext *s = ctx->priv;
     ThreadData *td = arg;
     FFTComplex *hdata = td->hdata;
     const int plane = td->plane;
     const int n = td->n;
caacbfa7
     int start = (n * jobnr) / nb_jobs;
c59e49f9
     int end = (n * (jobnr+1)) / nb_jobs;
f9f1b2a9
     int y;
 
c59e49f9
     for (y = start; y < end; y++) {
         av_fft_permute(s->fft[plane][jobnr], hdata + y * n);
         av_fft_calc(s->fft[plane][jobnr], hdata + y * n);
f9f1b2a9
     }
c59e49f9
 
     return 0;
f9f1b2a9
 }
 
 static void get_input(ConvolveContext *s, FFTComplex *fft_hdata,
                       AVFrame *in, int w, int h, int n, int plane, float scale)
4d41db7a
 {
5533cbfc
     const int iw = (n - w) / 2, ih = (n - h) / 2;
4d41db7a
     int y, x;
 
5533cbfc
     if (s->depth == 8) {
         for (y = 0; y < h; y++) {
4d41db7a
             const uint8_t *src = in->data[plane] + in->linesize[plane] * y;
 
             for (x = 0; x < w; x++) {
5533cbfc
                 fft_hdata[(y + ih) * n + iw + x].re = src[x] * scale;
ec6608f5
                 fft_hdata[(y + ih) * n + iw + x].im = 0;
4d41db7a
             }
 
5533cbfc
             for (x = 0; x < iw; x++) {
                 fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + iw].re;
ec6608f5
                 fft_hdata[(y + ih) * n + x].im = 0;
5533cbfc
             }
 
             for (x = n - iw; x < n; x++) {
                 fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + n - iw - 1].re;
ec6608f5
                 fft_hdata[(y + ih) * n + x].im = 0;
4d41db7a
             }
         }
b5958ff8
 
5533cbfc
         for (y = 0; y < ih; y++) {
             for (x = 0; x < n; x++) {
                 fft_hdata[y * n + x].re = fft_hdata[ih * n + x].re;
ec6608f5
                 fft_hdata[y * n + x].im = 0;
5533cbfc
             }
b5958ff8
         }
 
5533cbfc
         for (y = n - ih; y < n; y++) {
             for (x = 0; x < n; x++) {
                 fft_hdata[y * n + x].re = fft_hdata[(n - ih - 1) * n + x].re;
ec6608f5
                 fft_hdata[y * n + x].im = 0;
5533cbfc
             }
b5958ff8
         }
5533cbfc
     } else {
         for (y = 0; y < h; y++) {
             const uint16_t *src = (const uint16_t *)(in->data[plane] + in->linesize[plane] * y);
b5958ff8
 
5533cbfc
             for (x = 0; x < w; x++) {
                 fft_hdata[(y + ih) * n + iw + x].re = src[x] * scale;
ec6608f5
                 fft_hdata[(y + ih) * n + iw + x].im = 0;
5533cbfc
             }
 
             for (x = 0; x < iw; x++) {
                 fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + iw].re;
ec6608f5
                 fft_hdata[(y + ih) * n + x].im = 0;
5533cbfc
             }
 
             for (x = n - iw; x < n; x++) {
                 fft_hdata[(y + ih) * n + x].re = fft_hdata[(y + ih) * n + n - iw - 1].re;
ec6608f5
                 fft_hdata[(y + ih) * n + x].im = 0;
5533cbfc
             }
4d41db7a
         }
 
5533cbfc
         for (y = 0; y < ih; y++) {
             for (x = 0; x < n; x++) {
                 fft_hdata[y * n + x].re = fft_hdata[ih * n + x].re;
ec6608f5
                 fft_hdata[y * n + x].im = 0;
5533cbfc
             }
         }
 
         for (y = n - ih; y < n; y++) {
             for (x = 0; x < n; x++) {
                 fft_hdata[y * n + x].re = fft_hdata[(n - ih - 1) * n + x].re;
ec6608f5
                 fft_hdata[y * n + x].im = 0;
5533cbfc
             }
4d41db7a
         }
     }
 }
 
c59e49f9
 static int fft_vertical(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
4d41db7a
 {
c59e49f9
     ConvolveContext *s = ctx->priv;
     ThreadData *td = arg;
     FFTComplex *hdata = td->hdata;
     FFTComplex *vdata = td->vdata;
     const int plane = td->plane;
     const int n = td->n;
caacbfa7
     int start = (n * jobnr) / nb_jobs;
c59e49f9
     int end = (n * (jobnr+1)) / nb_jobs;
4d41db7a
     int y, x;
 
c59e49f9
     for (y = start; y < end; y++) {
4d41db7a
         for (x = 0; x < n; x++) {
c59e49f9
             vdata[y * n + x].re = hdata[x * n + y].re;
             vdata[y * n + x].im = hdata[x * n + y].im;
4d41db7a
         }
b5958ff8
 
c59e49f9
         av_fft_permute(s->fft[plane][jobnr], vdata + y * n);
         av_fft_calc(s->fft[plane][jobnr], vdata + y * n);
4d41db7a
     }
c59e49f9
 
     return 0;
4d41db7a
 }
 
c59e49f9
 static int ifft_vertical(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
4d41db7a
 {
c59e49f9
     ConvolveContext *s = ctx->priv;
     ThreadData *td = arg;
     FFTComplex *hdata = td->hdata;
     FFTComplex *vdata = td->vdata;
     const int plane = td->plane;
     const int n = td->n;
caacbfa7
     int start = (n * jobnr) / nb_jobs;
c59e49f9
     int end = (n * (jobnr+1)) / nb_jobs;
4d41db7a
     int y, x;
 
c59e49f9
     for (y = start; y < end; y++) {
         av_fft_permute(s->ifft[plane][jobnr], vdata + y * n);
         av_fft_calc(s->ifft[plane][jobnr], vdata + y * n);
b5958ff8
 
4d41db7a
         for (x = 0; x < n; x++) {
c59e49f9
             hdata[x * n + y].re = vdata[y * n + x].re;
             hdata[x * n + y].im = vdata[y * n + x].im;
4d41db7a
         }
     }
c59e49f9
 
     return 0;
4d41db7a
 }
 
c59e49f9
 static int ifft_horizontal(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
4d41db7a
 {
c59e49f9
     ConvolveContext *s = ctx->priv;
     ThreadData *td = arg;
     FFTComplex *hdata = td->hdata;
     const int plane = td->plane;
     const int n = td->n;
caacbfa7
     int start = (n * jobnr) / nb_jobs;
c59e49f9
     int end = (n * (jobnr+1)) / nb_jobs;
f9f1b2a9
     int y;
4d41db7a
 
c59e49f9
     for (y = start; y < end; y++) {
         av_fft_permute(s->ifft[plane][jobnr], hdata + y * n);
         av_fft_calc(s->ifft[plane][jobnr], hdata + y * n);
4d41db7a
     }
c59e49f9
 
     return 0;
f9f1b2a9
 }
 
8c9a91ac
 static void get_output(ConvolveContext *s, FFTComplex *input, AVFrame *out,
                        int w, int h, int n, int plane, float scale)
f9f1b2a9
 {
     const int max = (1 << s->depth) - 1;
     const int hh = h / 2;
     const int hw = w / 2;
     int y, x;
4d41db7a
 
     if (s->depth == 8) {
5533cbfc
         for (y = 0; y < hh; y++) {
             uint8_t *dst = out->data[plane] + (y + hh) * out->linesize[plane] + hw;
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip_uint8(input[y * n + x].re * scale);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
             uint8_t *dst = out->data[plane] + (y + hh) * out->linesize[plane];
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip_uint8(input[y * n + n - hw + x].re * scale);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
             uint8_t *dst = out->data[plane] + y * out->linesize[plane] + hw;
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip_uint8(input[(n - hh + y) * n + x].re * scale);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
4d41db7a
             uint8_t *dst = out->data[plane] + y * out->linesize[plane];
5533cbfc
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip_uint8(input[(n - hh + y) * n + n - hw + x].re * scale);
4d41db7a
         }
     } else {
5533cbfc
         for (y = 0; y < hh; y++) {
             uint16_t *dst = (uint16_t *)(out->data[plane] + (y + hh) * out->linesize[plane] + hw * 2);
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip(input[y * n + x].re * scale, 0, max);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
             uint16_t *dst = (uint16_t *)(out->data[plane] + (y + hh) * out->linesize[plane]);
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip(input[y * n + n - hw + x].re * scale, 0, max);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
             uint16_t *dst = (uint16_t *)(out->data[plane] + y * out->linesize[plane] + hw * 2);
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip(input[(n - hh + y) * n + x].re * scale, 0, max);
5533cbfc
         }
         for (y = 0; y < hh; y++) {
4d41db7a
             uint16_t *dst = (uint16_t *)(out->data[plane] + y * out->linesize[plane]);
5533cbfc
             for (x = 0; x < hw; x++)
86e3eebf
                 dst[x] = av_clip(input[(n - hh + y) * n + n - hw + x].re * scale, 0, max);
4d41db7a
         }
     }
 }
 
e712d301
 static int complex_multiply(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
8c9a91ac
     ConvolveContext *s = ctx->priv;
e712d301
     ThreadData *td = arg;
     FFTComplex *input = td->hdata;
     FFTComplex *filter = td->vdata;
8c9a91ac
     const float noise = s->noise;
e712d301
     const int n = td->n;
caacbfa7
     int start = (n * jobnr) / nb_jobs;
e712d301
     int end = (n * (jobnr+1)) / nb_jobs;
     int y, x;
 
     for (y = start; y < end; y++) {
         int yn = y * n;
 
         for (x = 0; x < n; x++) {
             FFTSample re, im, ire, iim;
 
             re = input[yn + x].re;
             im = input[yn + x].im;
8c9a91ac
             ire = filter[yn + x].re + noise;
e712d301
             iim = filter[yn + x].im;
 
             input[yn + x].re = ire * re - iim * im;
             input[yn + x].im = iim * re + ire * im;
         }
     }
 
     return 0;
 }
 
8c9a91ac
 static int complex_divide(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
     ConvolveContext *s = ctx->priv;
     ThreadData *td = arg;
     FFTComplex *input = td->hdata;
     FFTComplex *filter = td->vdata;
     const float noise = s->noise;
     const int n = td->n;
     int start = (n * jobnr) / nb_jobs;
     int end = (n * (jobnr+1)) / nb_jobs;
     int y, x;
 
     for (y = start; y < end; y++) {
         int yn = y * n;
 
         for (x = 0; x < n; x++) {
             FFTSample re, im, ire, iim, div;
 
             re = input[yn + x].re;
             im = input[yn + x].im;
             ire = filter[yn + x].re;
             iim = filter[yn + x].im;
             div = ire * ire + iim * iim + noise;
 
             input[yn + x].re = (ire * re + iim * im) / div;
             input[yn + x].im = (ire * im - iim * re) / div;
         }
     }
 
     return 0;
 }
 
4d41db7a
 static int do_convolve(FFFrameSync *fs)
 {
     AVFilterContext *ctx = fs->parent;
     AVFilterLink *outlink = ctx->outputs[0];
     ConvolveContext *s = ctx->priv;
     AVFrame *mainpic = NULL, *impulsepic = NULL;
     int ret, y, x, plane;
 
5f5dcf44
     ret = ff_framesync_dualinput_get(fs, &mainpic, &impulsepic);
4d41db7a
     if (ret < 0)
         return ret;
     if (!impulsepic)
         return ff_filter_frame(outlink, mainpic);
 
     for (plane = 0; plane < s->nb_planes; plane++) {
86e3eebf
         FFTComplex *filter = s->fft_vdata_impulse[plane];
         FFTComplex *input = s->fft_vdata[plane];
4d41db7a
         const int n = s->fft_len[plane];
         const int w = s->planewidth[plane];
         const int h = s->planeheight[plane];
         float total = 0;
c59e49f9
         ThreadData td;
4d41db7a
 
         if (!(s->planes & (1 << plane))) {
             continue;
         }
 
c59e49f9
         td.plane = plane, td.n = n;
f9f1b2a9
         get_input(s, s->fft_hdata[plane], mainpic, w, h, n, plane, 1.f);
c59e49f9
 
         td.hdata = s->fft_hdata[plane];
         td.vdata = s->fft_vdata[plane];
 
         ctx->internal->execute(ctx, fft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
         ctx->internal->execute(ctx, fft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
4d41db7a
 
         if ((!s->impulse && !s->got_impulse[plane]) || s->impulse) {
             if (s->depth == 8) {
                 for (y = 0; y < h; y++) {
                     const uint8_t *src = (const uint8_t *)(impulsepic->data[plane] + y * impulsepic->linesize[plane]) ;
                     for (x = 0; x < w; x++) {
                         total += src[x];
                     }
                 }
             } else {
                 for (y = 0; y < h; y++) {
                     const uint16_t *src = (const uint16_t *)(impulsepic->data[plane] + y * impulsepic->linesize[plane]) ;
                     for (x = 0; x < w; x++) {
                         total += src[x];
                     }
                 }
             }
             total = FFMAX(1, total);
 
8c9a91ac
             get_input(s, s->fft_hdata_impulse[plane], impulsepic, w, h, n, plane, 1.f / total);
c59e49f9
 
             td.hdata = s->fft_hdata_impulse[plane];
             td.vdata = s->fft_vdata_impulse[plane];
 
             ctx->internal->execute(ctx, fft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
             ctx->internal->execute(ctx, fft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
4d41db7a
 
             s->got_impulse[plane] = 1;
         }
 
e712d301
         td.hdata = input;
         td.vdata = filter;
86e3eebf
 
8c9a91ac
         ctx->internal->execute(ctx, s->filter, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
4d41db7a
 
c59e49f9
         td.hdata = s->fft_hdata[plane];
         td.vdata = s->fft_vdata[plane];
 
         ctx->internal->execute(ctx, ifft_vertical, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
         ctx->internal->execute(ctx, ifft_horizontal, &td, NULL, FFMIN3(MAX_THREADS, n, ff_filter_get_nb_threads(ctx)));
8c9a91ac
 
         get_output(s, s->fft_hdata[plane], mainpic, w, h, n, plane, 1.f / (n * n));
4d41db7a
     }
 
     return ff_filter_frame(outlink, mainpic);
 }
 
 static int config_output(AVFilterLink *outlink)
 {
     AVFilterContext *ctx = outlink->src;
     ConvolveContext *s = ctx->priv;
     AVFilterLink *mainlink = ctx->inputs[0];
c59e49f9
     int ret, i, j;
4d41db7a
 
     s->fs.on_event = do_convolve;
5f5dcf44
     ret = ff_framesync_init_dualinput(&s->fs, ctx);
4d41db7a
     if (ret < 0)
         return ret;
     outlink->w = mainlink->w;
     outlink->h = mainlink->h;
     outlink->time_base = mainlink->time_base;
     outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio;
     outlink->frame_rate = mainlink->frame_rate;
 
5f5dcf44
     if ((ret = ff_framesync_configure(&s->fs)) < 0)
4d41db7a
         return ret;
 
     for (i = 0; i < s->nb_planes; i++) {
c59e49f9
         for (j = 0; j < MAX_THREADS; j++) {
             s->fft[i][j]  = av_fft_init(s->fft_bits[i], 0);
             s->ifft[i][j] = av_fft_init(s->fft_bits[i], 1);
             if (!s->fft[i][j] || !s->ifft[i][j])
                 return AVERROR(ENOMEM);
         }
4d41db7a
     }
 
     return 0;
 }
 
 static int activate(AVFilterContext *ctx)
 {
     ConvolveContext *s = ctx->priv;
5f5dcf44
     return ff_framesync_activate(&s->fs);
4d41db7a
 }
 
8c9a91ac
 static av_cold int init(AVFilterContext *ctx)
 {
     ConvolveContext *s = ctx->priv;
 
     if (!strcmp(ctx->filter->name, "convolve")) {
         s->filter = complex_multiply;
     } else if (!strcmp(ctx->filter->name, "deconvolve")) {
         s->filter = complex_divide;
     } else {
         return AVERROR_BUG;
     }
 
     return 0;
 }
 
4d41db7a
 static av_cold void uninit(AVFilterContext *ctx)
 {
     ConvolveContext *s = ctx->priv;
c59e49f9
     int i, j;
4d41db7a
 
     for (i = 0; i < 4; i++) {
         av_freep(&s->fft_hdata[i]);
         av_freep(&s->fft_vdata[i]);
         av_freep(&s->fft_hdata_impulse[i]);
         av_freep(&s->fft_vdata_impulse[i]);
c59e49f9
 
         for (j = 0; j < MAX_THREADS; j++) {
             av_fft_end(s->fft[i][j]);
             av_fft_end(s->ifft[i][j]);
         }
4d41db7a
     }
 
5f5dcf44
     ff_framesync_uninit(&s->fs);
4d41db7a
 }
 
 static const AVFilterPad convolve_inputs[] = {
     {
         .name          = "main",
         .type          = AVMEDIA_TYPE_VIDEO,
         .config_props  = config_input_main,
     },{
         .name          = "impulse",
         .type          = AVMEDIA_TYPE_VIDEO,
         .config_props  = config_input_impulse,
     },
     { NULL }
 };
 
 static const AVFilterPad convolve_outputs[] = {
     {
         .name          = "default",
         .type          = AVMEDIA_TYPE_VIDEO,
         .config_props  = config_output,
     },
     { NULL }
 };
 
8c9a91ac
 #if CONFIG_CONVOLVE_FILTER
 
 FRAMESYNC_DEFINE_CLASS(convolve, ConvolveContext, fs);
 
4d41db7a
 AVFilter ff_vf_convolve = {
     .name          = "convolve",
     .description   = NULL_IF_CONFIG_SMALL("Convolve first video stream with second video stream."),
     .preinit       = convolve_framesync_preinit,
8c9a91ac
     .init          = init,
4d41db7a
     .uninit        = uninit,
     .query_formats = query_formats,
     .activate      = activate,
     .priv_size     = sizeof(ConvolveContext),
     .priv_class    = &convolve_class,
     .inputs        = convolve_inputs,
     .outputs       = convolve_outputs,
c59e49f9
     .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
4d41db7a
 };
8c9a91ac
 
 #endif /* CONFIG_CONVOLVE_FILTER */
 
 #if CONFIG_DECONVOLVE_FILTER
 
 static const AVOption deconvolve_options[] = {
     { "planes",  "set planes to deconvolve",                OFFSET(planes),   AV_OPT_TYPE_INT,   {.i64=7}, 0, 15, FLAGS },
     { "impulse", "when to process impulses",                OFFSET(impulse),  AV_OPT_TYPE_INT,   {.i64=1}, 0,  1, FLAGS, "impulse" },
     {   "first", "process only first impulse, ignore rest", 0,                AV_OPT_TYPE_CONST, {.i64=0}, 0,  0, FLAGS, "impulse" },
     {   "all",   "process all impulses",                    0,                AV_OPT_TYPE_CONST, {.i64=1}, 0,  0, FLAGS, "impulse" },
     { "noise",   "set noise",                               OFFSET(noise),    AV_OPT_TYPE_FLOAT, {.dbl=0.0000001}, 0,  1, FLAGS },
     { NULL },
 };
 
 FRAMESYNC_DEFINE_CLASS(deconvolve, ConvolveContext, fs);
 
 AVFilter ff_vf_deconvolve = {
     .name          = "deconvolve",
     .description   = NULL_IF_CONFIG_SMALL("Deconvolve first video stream with second video stream."),
     .preinit       = deconvolve_framesync_preinit,
     .init          = init,
     .uninit        = uninit,
     .query_formats = query_formats,
     .activate      = activate,
     .priv_size     = sizeof(ConvolveContext),
     .priv_class    = &deconvolve_class,
     .inputs        = convolve_inputs,
     .outputs       = convolve_outputs,
     .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
 };
 
 #endif /* CONFIG_DECONVOLVE_FILTER */