2 * Copyright (c) 2011 Pascal Getreuer
3 * Copyright (c) 2016 Paul B Mahol
5 * Redistribution and use in source and binary forms, with or without modification,
6 * are permitted provided that the following conditions are met:
8 * * Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above
11 * copyright notice, this list of conditions and the following
12 * disclaimer in the documentation and/or other materials provided
13 * with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
19 * HOLDER BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
20 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
21 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
22 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
23 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
24 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 #include "libavutil/imgutils.h"
29 #include "libavutil/opt.h"
30 #include "libavutil/pixdesc.h"
37 #define OFFSET(x) offsetof(GBlurContext, x)
38 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
40 static const AVOption gblur_options[] = {
41 { "sigma", "set sigma", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl=0.5}, 0.0, 1024, FLAGS },
42 { "steps", "set number of steps", OFFSET(steps), AV_OPT_TYPE_INT, {.i64=1}, 1, 6, FLAGS },
43 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=0xF}, 0, 0xF, FLAGS },
44 { "sigmaV", "set vertical sigma", OFFSET(sigmaV), AV_OPT_TYPE_FLOAT, {.dbl=-1}, -1, 1024, FLAGS },
48 AVFILTER_DEFINE_CLASS(gblur);
50 typedef struct ThreadData {
55 static void horiz_slice_c(float *buffer, int width, int height, int steps,
56 float nu, float bscale)
60 for (y = 0; y < height; y++) {
61 for (step = 0; step < steps; step++) {
62 ptr = buffer + width * y;
65 /* Filter rightwards */
66 for (x = 1; x < width; x++)
67 ptr[x] += nu * ptr[x - 1];
68 ptr[x = width - 1] *= bscale;
70 /* Filter leftwards */
72 ptr[x - 1] += nu * ptr[x];
77 static int filter_horizontally(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
79 GBlurContext *s = ctx->priv;
81 const int height = td->height;
82 const int width = td->width;
83 const int slice_start = (height * jobnr ) / nb_jobs;
84 const int slice_end = (height * (jobnr+1)) / nb_jobs;
85 const float boundaryscale = s->boundaryscale;
86 const int steps = s->steps;
87 const float nu = s->nu;
88 float *buffer = s->buffer;
90 s->horiz_slice(buffer + width * slice_start, width, slice_end - slice_start,
91 steps, nu, boundaryscale);
96 static void do_vertical_columns(float *buffer, int width, int height,
97 int column_begin, int column_end, int steps,
98 float nu, float boundaryscale, int column_step)
100 const int numpixels = width * height;
103 for (x = column_begin; x < column_end;) {
104 for (step = 0; step < steps; step++) {
106 for (k = 0; k < column_step; k++) {
107 ptr[k] *= boundaryscale;
109 /* Filter downwards */
110 for (i = width; i < numpixels; i += width) {
111 for (k = 0; k < column_step; k++) {
112 ptr[i + k] += nu * ptr[i - width + k];
115 i = numpixels - width;
117 for (k = 0; k < column_step; k++)
118 ptr[i + k] *= boundaryscale;
121 for (; i > 0; i -= width) {
122 for (k = 0; k < column_step; k++)
123 ptr[i - width + k] += nu * ptr[i + k];
130 static int filter_vertically(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
132 GBlurContext *s = ctx->priv;
133 ThreadData *td = arg;
134 const int height = td->height;
135 const int width = td->width;
136 const int slice_start = (width * jobnr ) / nb_jobs;
137 const int slice_end = (width * (jobnr+1)) / nb_jobs;
138 const float boundaryscale = s->boundaryscaleV;
139 const int steps = s->steps;
140 const float nu = s->nuV;
141 float *buffer = s->buffer;
144 aligned_end = slice_start + (((slice_end - slice_start) >> 3) << 3);
145 /* Filter vertically along columns (process 8 columns in each step) */
146 do_vertical_columns(buffer, width, height, slice_start, aligned_end,
147 steps, nu, boundaryscale, 8);
149 /* Filter un-aligned columns one by one */
150 do_vertical_columns(buffer, width, height, aligned_end, slice_end,
151 steps, nu, boundaryscale, 1);
156 static int filter_postscale(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
158 GBlurContext *s = ctx->priv;
159 ThreadData *td = arg;
160 const int height = td->height;
161 const int width = td->width;
162 const int64_t numpixels = width * (int64_t)height;
163 const unsigned slice_start = (numpixels * jobnr ) / nb_jobs;
164 const unsigned slice_end = (numpixels * (jobnr+1)) / nb_jobs;
165 const float postscale = s->postscale * s->postscaleV;
166 float *buffer = s->buffer;
169 for (i = slice_start; i < slice_end; i++)
170 buffer[i] *= postscale;
175 static void gaussianiir2d(AVFilterContext *ctx, int plane)
177 GBlurContext *s = ctx->priv;
178 const int width = s->planewidth[plane];
179 const int height = s->planeheight[plane];
180 const int nb_threads = ff_filter_get_nb_threads(ctx);
183 if (s->sigma <= 0 || s->steps < 0)
188 ctx->internal->execute(ctx, filter_horizontally, &td, NULL, FFMIN(height, nb_threads));
189 ctx->internal->execute(ctx, filter_vertically, &td, NULL, FFMIN(width, nb_threads));
190 ctx->internal->execute(ctx, filter_postscale, &td, NULL, FFMIN(width * height, nb_threads));
193 static int query_formats(AVFilterContext *ctx)
195 static const enum AVPixelFormat pix_fmts[] = {
196 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
197 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
198 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
199 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
200 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
201 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
202 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
203 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
204 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
205 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
206 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
207 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
208 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
209 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
210 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
211 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
212 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
216 return ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
219 void ff_gblur_init(GBlurContext *s)
221 s->horiz_slice = horiz_slice_c;
223 ff_gblur_init_x86(s);
226 static int config_input(AVFilterLink *inlink)
228 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
229 GBlurContext *s = inlink->dst->priv;
231 s->depth = desc->comp[0].depth;
232 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
233 s->planewidth[0] = s->planewidth[3] = inlink->w;
234 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
235 s->planeheight[0] = s->planeheight[3] = inlink->h;
237 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
239 s->buffer = av_malloc_array(inlink->w, inlink->h * sizeof(*s->buffer));
241 return AVERROR(ENOMEM);
244 s->sigmaV = s->sigma;
251 static void set_params(float sigma, int steps, float *postscale, float *boundaryscale, float *nu)
255 lambda = (sigma * sigma) / (2.0 * steps);
256 dnu = (1.0 + 2.0 * lambda - sqrt(1.0 + 4.0 * lambda)) / (2.0 * lambda);
257 *postscale = pow(dnu / lambda, steps);
258 *boundaryscale = 1.0 / (1.0 - dnu);
262 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
264 AVFilterContext *ctx = inlink->dst;
265 GBlurContext *s = ctx->priv;
266 AVFilterLink *outlink = ctx->outputs[0];
270 set_params(s->sigma, s->steps, &s->postscale, &s->boundaryscale, &s->nu);
271 set_params(s->sigmaV, s->steps, &s->postscaleV, &s->boundaryscaleV, &s->nuV);
273 if (av_frame_is_writable(in)) {
276 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
279 return AVERROR(ENOMEM);
281 av_frame_copy_props(out, in);
284 for (plane = 0; plane < s->nb_planes; plane++) {
285 const int height = s->planeheight[plane];
286 const int width = s->planewidth[plane];
287 float *bptr = s->buffer;
288 const uint8_t *src = in->data[plane];
289 const uint16_t *src16 = (const uint16_t *)in->data[plane];
290 uint8_t *dst = out->data[plane];
291 uint16_t *dst16 = (uint16_t *)out->data[plane];
294 if (!s->sigma || !(s->planes & (1 << plane))) {
296 av_image_copy_plane(out->data[plane], out->linesize[plane],
297 in->data[plane], in->linesize[plane],
298 width * ((s->depth + 7) / 8), height);
303 for (y = 0; y < height; y++) {
304 for (x = 0; x < width; x++) {
308 src += in->linesize[plane];
311 for (y = 0; y < height; y++) {
312 for (x = 0; x < width; x++) {
316 src16 += in->linesize[plane] / 2;
320 gaussianiir2d(ctx, plane);
324 for (y = 0; y < height; y++) {
325 for (x = 0; x < width; x++) {
329 dst += out->linesize[plane];
332 for (y = 0; y < height; y++) {
333 for (x = 0; x < width; x++) {
337 dst16 += out->linesize[plane] / 2;
344 return ff_filter_frame(outlink, out);
347 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
348 char *res, int res_len, int flags)
350 GBlurContext *s = ctx->priv;
353 if ( !strcmp(cmd, "sigma") || !strcmp(cmd, "sigmaV")
354 || !strcmp(cmd, "steps") || !strcmp(cmd, "planes")) {
355 av_opt_set(s, cmd, args, 0);
357 ret = AVERROR(ENOSYS);
363 static av_cold void uninit(AVFilterContext *ctx)
365 GBlurContext *s = ctx->priv;
367 av_freep(&s->buffer);
370 static const AVFilterPad gblur_inputs[] = {
373 .type = AVMEDIA_TYPE_VIDEO,
374 .config_props = config_input,
375 .filter_frame = filter_frame,
380 static const AVFilterPad gblur_outputs[] = {
383 .type = AVMEDIA_TYPE_VIDEO,
388 AVFilter ff_vf_gblur = {
390 .description = NULL_IF_CONFIG_SMALL("Apply Gaussian Blur filter."),
391 .priv_size = sizeof(GBlurContext),
392 .priv_class = &gblur_class,
394 .query_formats = query_formats,
395 .inputs = gblur_inputs,
396 .outputs = gblur_outputs,
397 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
398 .process_command = process_command,