2 * This file is part of FFmpeg.
4 * Copyright (c) 2011, 2012 Hyllian/Jararaca <sergiogdb@gmail.com>
5 * Copyright (c) 2014 Arwa Arif <arwaarif1994@gmail.com>
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
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 GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * XBR Filter is used for depixelization of image.
25 * This is based on Hyllian's xBR shader.
27 * @see http://www.libretro.com/forums/viewtopic.php?f=6&t=134
28 * @see https://github.com/yoyofr/iFBA/blob/master/fba_src/src/intf/video/scalers/xbr.cpp
31 #include "libavutil/opt.h"
32 #include "libavutil/avassert.h"
33 #include "libavutil/pixdesc.h"
36 #define RGB_MASK 0x00FFFFFF
37 #define LB_MASK 0x00FEFEFE
38 #define RED_BLUE_MASK 0x00FF00FF
39 #define GREEN_MASK 0x0000FF00
41 typedef int (*xbrfunc_t)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
47 uint32_t rgbtoyuv[1<<24];
50 typedef struct ThreadData {
52 const uint32_t *rgbtoyuv;
55 #define OFFSET(x) offsetof(XBRContext, x)
56 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
57 static const AVOption xbr_options[] = {
58 { "n", "set scale factor", OFFSET(n), AV_OPT_TYPE_INT, {.i64 = 3}, 2, 4, .flags = FLAGS },
62 AVFILTER_DEFINE_CLASS(xbr);
64 static uint32_t pixel_diff(uint32_t x, uint32_t y, const uint32_t *r2y)
66 #define YMASK 0xff0000
67 #define UMASK 0x00ff00
68 #define VMASK 0x0000ff
70 uint32_t yuv1 = r2y[x & 0xffffff];
71 uint32_t yuv2 = r2y[y & 0xffffff];
73 return (abs((yuv1 & YMASK) - (yuv2 & YMASK)) >> 16) +
74 (abs((yuv1 & UMASK) - (yuv2 & UMASK)) >> 8) +
75 abs((yuv1 & VMASK) - (yuv2 & VMASK));
78 #define ALPHA_BLEND_128_W(dst, src) dst = ((src & LB_MASK) >> 1) + ((dst & LB_MASK) >> 1)
80 #define ALPHA_BLEND_32_W(dst, src) \
81 dst = ((RED_BLUE_MASK & ((dst & RED_BLUE_MASK) + ((((src & RED_BLUE_MASK) - \
82 (dst & RED_BLUE_MASK))) >> 3))) | (GREEN_MASK & ((dst & GREEN_MASK) + \
83 ((((src & GREEN_MASK) - (dst & GREEN_MASK))) >> 3))))
85 #define ALPHA_BLEND_64_W(dst, src) \
86 dst = ((RED_BLUE_MASK & ((dst & RED_BLUE_MASK) + ((((src & RED_BLUE_MASK) - \
87 (dst & RED_BLUE_MASK))) >> 2))) | (GREEN_MASK & ((dst & GREEN_MASK) + \
88 ((((src & GREEN_MASK) - (dst & GREEN_MASK))) >> 2))))
90 #define ALPHA_BLEND_192_W(dst, src) \
91 dst = ((RED_BLUE_MASK & ((dst & RED_BLUE_MASK) + ((((src & RED_BLUE_MASK) - \
92 (dst & RED_BLUE_MASK)) * 3) >> 2))) | (GREEN_MASK & ((dst & GREEN_MASK) + \
93 ((((src & GREEN_MASK) - (dst & GREEN_MASK)) * 3) >> 2))))
95 #define ALPHA_BLEND_224_W(dst, src) \
96 dst = ((RED_BLUE_MASK & ((dst & RED_BLUE_MASK) + ((((src & RED_BLUE_MASK) - \
97 (dst & RED_BLUE_MASK)) * 7) >> 3))) | (GREEN_MASK & ((dst & GREEN_MASK) + \
98 ((((src & GREEN_MASK) - (dst & GREEN_MASK)) * 7) >> 3))))
100 #define df(A, B) pixel_diff(A, B, r2y)
101 #define eq(A, B) (df(A, B) < 155)
103 #define FILT2(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
104 N0, N1, N2, N3) do { \
105 if (PE != PH && PE != PF) { \
106 const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
107 const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
108 if (e < i && (!eq(PF,PB) && !eq(PH,PD) || eq(PE,PI) \
109 && (!eq(PF,I4) && !eq(PH,I5)) \
110 || eq(PE,PG) || eq(PE,PC))) { \
111 const unsigned ke = df(PF,PG); \
112 const unsigned ki = df(PH,PC); \
113 const int left = ke<<1 <= ki && PE != PG && PD != PG; \
114 const int up = ke >= ki<<1 && PE != PC && PB != PC; \
115 const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
117 ALPHA_BLEND_224_W(E[N3], px); \
118 ALPHA_BLEND_64_W( E[N2], px); \
121 ALPHA_BLEND_192_W(E[N3], px); \
122 ALPHA_BLEND_64_W( E[N2], px); \
124 ALPHA_BLEND_192_W(E[N3], px); \
125 ALPHA_BLEND_64_W( E[N1], px); \
126 } else { /* diagonal */ \
127 ALPHA_BLEND_128_W(E[N3], px); \
129 } else if (e <= i) { \
130 ALPHA_BLEND_128_W( E[N3], ((df(PE,PF) <= df(PE,PH)) ? PF : PH)); \
135 #define FILT3(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
136 N0, N1, N2, N3, N4, N5, N6, N7, N8) do { \
137 if (PE != PH && PE != PF) { \
138 const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
139 const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
140 if (e < i && (!eq(PF,PB) && !eq(PF,PC) || !eq(PH,PD) && !eq(PH,PG) || eq(PE,PI) \
141 && (!eq(PF,F4) && !eq(PF,I4) || !eq(PH,H5) && !eq(PH,I5)) \
142 || eq(PE,PG) || eq(PE,PC))) { \
143 const unsigned ke = df(PF,PG); \
144 const unsigned ki = df(PH,PC); \
145 const int left = ke<<1 <= ki && PE != PG && PD != PG; \
146 const int up = ke >= ki<<1 && PE != PC && PB != PC; \
147 const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
149 ALPHA_BLEND_192_W(E[N7], px); \
150 ALPHA_BLEND_64_W( E[N6], px); \
155 ALPHA_BLEND_192_W(E[N7], px); \
156 ALPHA_BLEND_64_W( E[N5], px); \
157 ALPHA_BLEND_64_W( E[N6], px); \
160 ALPHA_BLEND_192_W(E[N5], px); \
161 ALPHA_BLEND_64_W( E[N7], px); \
162 ALPHA_BLEND_64_W( E[N2], px); \
164 } else { /* diagonal */ \
165 ALPHA_BLEND_224_W(E[N8], px); \
166 ALPHA_BLEND_32_W( E[N5], px); \
167 ALPHA_BLEND_32_W( E[N7], px); \
169 } else if (e <= i) { \
170 ALPHA_BLEND_128_W(E[N8], ((df(PE,PF) <= df(PE,PH)) ? PF : PH)); \
175 #define FILT4(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, \
176 N15, N14, N11, N3, N7, N10, N13, N12, N9, N6, N2, N1, N5, N8, N4, N0) do { \
177 if (PE != PH && PE != PF) { \
178 const unsigned e = df(PE,PC) + df(PE,PG) + df(PI,H5) + df(PI,F4) + (df(PH,PF)<<2); \
179 const unsigned i = df(PH,PD) + df(PH,I5) + df(PF,I4) + df(PF,PB) + (df(PE,PI)<<2); \
180 if (e < i && (!eq(PF,PB) && !eq(PH,PD) || eq(PE,PI) \
181 && (!eq(PF,I4) && !eq(PH,I5)) \
182 || eq(PE,PG) || eq(PE,PC))) { \
183 const unsigned ke = df(PF,PG); \
184 const unsigned ki = df(PH,PC); \
185 const int left = ke<<1 <= ki && PE != PG && PD != PG; \
186 const int up = ke >= ki<<1 && PE != PC && PB != PC; \
187 const unsigned px = df(PE,PF) <= df(PE,PH) ? PF : PH; \
189 ALPHA_BLEND_192_W(E[N13], px); \
190 ALPHA_BLEND_64_W( E[N12], px); \
191 E[N15] = E[N14] = E[N11] = px; \
192 E[N10] = E[N3] = E[N12]; \
195 ALPHA_BLEND_192_W(E[N11], px); \
196 ALPHA_BLEND_192_W(E[N13], px); \
197 ALPHA_BLEND_64_W( E[N10], px); \
198 ALPHA_BLEND_64_W( E[N12], px); \
202 ALPHA_BLEND_192_W(E[N14], px); \
203 ALPHA_BLEND_192_W(E[N7 ], px); \
204 ALPHA_BLEND_64_W( E[N10], px); \
205 ALPHA_BLEND_64_W( E[N3 ], px); \
208 } else { /* diagonal */ \
209 ALPHA_BLEND_128_W(E[N11], px); \
210 ALPHA_BLEND_128_W(E[N14], px); \
213 } else if (e <= i) { \
214 ALPHA_BLEND_128_W( E[N15], ((df(PE,PF) <= df(PE,PH)) ? PF : PH)); \
219 static av_always_inline void xbr_filter(const ThreadData *td, int jobnr, int nb_jobs, int n)
222 const AVFrame *input = td->in;
223 AVFrame *output = td->out;
224 const uint32_t *r2y = td->rgbtoyuv;
225 const int slice_start = (input->height * jobnr ) / nb_jobs;
226 const int slice_end = (input->height * (jobnr+1)) / nb_jobs;
227 const int nl = output->linesize[0] >> 2;
228 const int nl1 = nl + nl;
229 const int nl2 = nl1 + nl;
231 for (y = slice_start; y < slice_end; y++) {
233 uint32_t *E = (uint32_t *)(output->data[0] + y * output->linesize[0] * n);
234 const uint32_t *sa2 = (uint32_t *)(input->data[0] + y * input->linesize[0] - 8); /* center */
235 const uint32_t *sa1 = sa2 - (input->linesize[0]>>2); /* up x1 */
236 const uint32_t *sa0 = sa1 - (input->linesize[0]>>2); /* up x2 */
237 const uint32_t *sa3 = sa2 + (input->linesize[0]>>2); /* down x1 */
238 const uint32_t *sa4 = sa3 + (input->linesize[0]>>2); /* down x2 */
247 if (y >= input->height - 2) {
249 if (y == input->height - 1) {
254 for (x = 0; x < input->width; x++) {
255 const uint32_t B1 = sa0[2];
256 const uint32_t PB = sa1[2];
257 const uint32_t PE = sa2[2];
258 const uint32_t PH = sa3[2];
259 const uint32_t H5 = sa4[2];
261 const int pprev = 2 - (x > 0);
262 const uint32_t A1 = sa0[pprev];
263 const uint32_t PA = sa1[pprev];
264 const uint32_t PD = sa2[pprev];
265 const uint32_t PG = sa3[pprev];
266 const uint32_t G5 = sa4[pprev];
268 const int pprev2 = pprev - (x > 1);
269 const uint32_t A0 = sa1[pprev2];
270 const uint32_t D0 = sa2[pprev2];
271 const uint32_t G0 = sa3[pprev2];
273 const int pnext = 3 - (x == input->width - 1);
274 const uint32_t C1 = sa0[pnext];
275 const uint32_t PC = sa1[pnext];
276 const uint32_t PF = sa2[pnext];
277 const uint32_t PI = sa3[pnext];
278 const uint32_t I5 = sa4[pnext];
280 const int pnext2 = pnext + 1 - (x >= input->width - 2);
281 const uint32_t C4 = sa1[pnext2];
282 const uint32_t F4 = sa2[pnext2];
283 const uint32_t I4 = sa3[pnext2];
286 E[0] = E[1] = // 0, 1
287 E[nl] = E[nl + 1] = PE; // 2, 3
289 FILT2(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, 0, 1, nl, nl+1);
290 FILT2(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, nl, 0, nl+1, 1);
291 FILT2(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, nl+1, nl, 1, 0);
292 FILT2(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, 1, nl+1, 0, nl);
294 E[0] = E[1] = E[2] = // 0, 1, 2
295 E[nl] = E[nl+1] = E[nl+2] = // 3, 4, 5
296 E[nl1] = E[nl1+1] = E[nl1+2] = PE; // 6, 7, 8
298 FILT3(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, 0, 1, 2, nl, nl+1, nl+2, nl1, nl1+1, nl1+2);
299 FILT3(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, nl1, nl, 0, nl1+1, nl+1, 1, nl1+2, nl+2, 2);
300 FILT3(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, nl1+2, nl1+1, nl1, nl+2, nl+1, nl, 2, 1, 0);
301 FILT3(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, 2, nl+2, nl1+2, 1, nl+1, nl1+1, 0, nl, nl1);
303 E[0] = E[1] = E[2] = E[3] = // 0, 1, 2, 3
304 E[nl] = E[nl+1] = E[nl+2] = E[nl+3] = // 4, 5, 6, 7
305 E[nl1] = E[nl1+1] = E[nl1+2] = E[nl1+3] = // 8, 9, 10, 11
306 E[nl2] = E[nl2+1] = E[nl2+2] = E[nl2+3] = PE; // 12, 13, 14, 15
308 FILT4(PE, PI, PH, PF, PG, PC, PD, PB, PA, G5, C4, G0, D0, C1, B1, F4, I4, H5, I5, A0, A1, nl2+3, nl2+2, nl1+3, 3, nl+3, nl1+2, nl2+1, nl2, nl1+1, nl+2, 2, 1, nl+1, nl1, nl, 0);
309 FILT4(PE, PC, PF, PB, PI, PA, PH, PD, PG, I4, A1, I5, H5, A0, D0, B1, C1, F4, C4, G5, G0, 3, nl+3, 2, 0, 1, nl+2, nl1+3, nl2+3, nl1+2, nl+1, nl, nl1, nl1+1,nl2+2,nl2+1,nl2);
310 FILT4(PE, PA, PB, PD, PC, PG, PF, PH, PI, C1, G0, C4, F4, G5, H5, D0, A0, B1, A1, I4, I5, 0, 1, nl, nl2, nl1, nl+1, 2, 3, nl+2, nl1+1, nl2+1,nl2+2,nl1+2, nl+3,nl1+3,nl2+3);
311 FILT4(PE, PG, PD, PH, PA, PI, PB, PF, PC, A0, I5, A1, B1, I4, F4, H5, G5, D0, G0, C1, C4, nl2, nl1, nl2+1, nl2+3, nl2+2, nl1+1, nl, 0, nl+1, nl1+2, nl1+3, nl+3, nl+2, 1, 2, 3);
325 #define XBR_FUNC(size) \
326 static int xbr##size##x(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
328 xbr_filter(arg, jobnr, nb_jobs, size); \
337 static int config_output(AVFilterLink *outlink)
339 AVFilterContext *ctx = outlink->src;
340 XBRContext *xbr = ctx->priv;
341 AVFilterLink *inlink = ctx->inputs[0];
343 outlink->w = inlink->w * xbr->n;
344 outlink->h = inlink->h * xbr->n;
348 static int query_formats(AVFilterContext *ctx)
350 static const enum AVPixelFormat pix_fmts[] = {
351 AV_PIX_FMT_0RGB32, AV_PIX_FMT_NONE,
354 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
358 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
360 AVFilterContext *ctx = inlink->dst;
361 AVFilterLink *outlink = ctx->outputs[0];
362 XBRContext *xbr = ctx->priv;
365 AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
368 return AVERROR(ENOMEM);
371 av_frame_copy_props(out, in);
375 td.rgbtoyuv = xbr->rgbtoyuv;
376 ctx->internal->execute(ctx, xbr->func, &td, NULL, FFMIN(inlink->h, ctx->graph->nb_threads));
378 out->width = outlink->w;
379 out->height = outlink->h;
382 return ff_filter_frame(outlink, out);
385 static int init(AVFilterContext *ctx)
387 XBRContext *xbr = ctx->priv;
388 static const xbrfunc_t xbrfuncs[] = {xbr2x, xbr3x, xbr4x};
393 for (bg = -255; bg < 256; bg++) {
394 for (rg = -255; rg < 256; rg++) {
395 const uint32_t u = (uint32_t)((-169*rg + 500*bg)/1000) + 128;
396 const uint32_t v = (uint32_t)(( 500*rg - 81*bg)/1000) + 128;
397 int startg = FFMAX3(-bg, -rg, 0);
398 int endg = FFMIN3(255-bg, 255-rg, 255);
399 uint32_t y = (uint32_t)(( 299*rg + 1000*startg + 114*bg)/1000);
400 c = bg + (rg<<16) + 0x010101 * startg;
401 for (g = startg; g <= endg; g++) {
402 xbr->rgbtoyuv[c] = ((y++) << 16) + (u << 8) + v;
408 xbr->func = xbrfuncs[xbr->n - 2];
412 static const AVFilterPad xbr_inputs[] = {
415 .type = AVMEDIA_TYPE_VIDEO,
416 .filter_frame = filter_frame,
421 static const AVFilterPad xbr_outputs[] = {
424 .type = AVMEDIA_TYPE_VIDEO,
425 .config_props = config_output,
430 AVFilter ff_vf_xbr = {
432 .description = NULL_IF_CONFIG_SMALL("Scale the input using xBR algorithm."),
433 .inputs = xbr_inputs,
434 .outputs = xbr_outputs,
435 .query_formats = query_formats,
436 .priv_size = sizeof(XBRContext),
437 .priv_class = &xbr_class,
439 .flags = AVFILTER_FLAG_SLICE_THREADS,