1 /*
2 * Copyright (c) 2014 Clément Bœsch
3 *
4 * This file is part of FFmpeg.
5 *
6 * Permission to use, copy, modify, and/or distribute this software for any
7 * purpose with or without fee is hereby granted, provided that the above
8 * copyright notice and this permission notice appear in all copies.
9 *
10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17 */
18
19 /**
20 * @file
21 * hqx magnification filters (hq2x, hq3x, hq4x)
22 *
23 * Originally designed by Maxim Stephin.
24 *
25 * @see http://en.wikipedia.org/wiki/Hqx
26 * @see http://web.archive.org/web/20131114143602/http://www.hiend3d.com/hq3x.html
27 * @see http://blog.pkh.me/p/19-butchering-hqx-scaling-filters.html
28 */
29
30 #include "libavutil/opt.h"
31 #include "libavutil/avassert.h"
32 #include "libavutil/pixdesc.h"
33 #include "internal.h"
34
35 typedef int (*hqxfunc_t)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
36
37 typedef struct HQXContext {
38 const AVClass *class;
39 int n;
40 hqxfunc_t func;
41 uint32_t rgbtoyuv[1<<24];
42 } HQXContext;
43
44 typedef struct ThreadData {
45 AVFrame *in, *out;
46 const uint32_t *rgbtoyuv;
47 } ThreadData;
48
49 #define OFFSET(x) offsetof(HQXContext, x)
50 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
51 static const AVOption hqx_options[] = {
52 { "n", "set scale factor", OFFSET(n), AV_OPT_TYPE_INT, {.i64 = 3}, 2, 4, .flags = FLAGS },
53 { NULL }
54 };
55
56 AVFILTER_DEFINE_CLASS(hqx);
57
rgb2yuv(const uint32_t * r2y,uint32_t c)58 static av_always_inline uint32_t rgb2yuv(const uint32_t *r2y, uint32_t c)
59 {
60 return r2y[c & 0xffffff];
61 }
62
yuv_diff(uint32_t yuv1,uint32_t yuv2)63 static av_always_inline int yuv_diff(uint32_t yuv1, uint32_t yuv2)
64 {
65 #define YMASK 0xff0000
66 #define UMASK 0x00ff00
67 #define VMASK 0x0000ff
68 #define ABSDIFF(a,b) (abs((int)(a)-(int)(b)))
69
70 return ABSDIFF(yuv1 & YMASK, yuv2 & YMASK) > (48 << 16) ||
71 ABSDIFF(yuv1 & UMASK, yuv2 & UMASK) > ( 7 << 8) ||
72 ABSDIFF(yuv1 & VMASK, yuv2 & VMASK) > ( 6 << 0);
73 }
74
75 /* (c1*w1 + c2*w2) >> s */
interp_2px(uint32_t c1,int w1,uint32_t c2,int w2,int s)76 static av_always_inline uint32_t interp_2px(uint32_t c1, int w1, uint32_t c2, int w2, int s)
77 {
78 return (((((c1 & 0xff00ff00) >> 8) * w1 + ((c2 & 0xff00ff00) >> 8) * w2) << (8 - s)) & 0xff00ff00) |
79 (((((c1 & 0x00ff00ff) ) * w1 + ((c2 & 0x00ff00ff) ) * w2) >> s ) & 0x00ff00ff);
80 }
81
82 /* (c1*w1 + c2*w2 + c3*w3) >> s */
interp_3px(uint32_t c1,int w1,uint32_t c2,int w2,uint32_t c3,int w3,int s)83 static av_always_inline uint32_t interp_3px(uint32_t c1, int w1, uint32_t c2, int w2, uint32_t c3, int w3, int s)
84 {
85 return (((((c1 & 0xff00ff00) >> 8) * w1 + ((c2 & 0xff00ff00) >> 8) * w2 + ((c3 & 0xff00ff00) >> 8) * w3) << (8 - s)) & 0xff00ff00) |
86 (((((c1 & 0x00ff00ff) ) * w1 + ((c2 & 0x00ff00ff) ) * w2 + ((c3 & 0x00ff00ff) ) * w3) >> s ) & 0x00ff00ff);
87 }
88
89 /* m is the mask of diff with the center pixel that matters in the pattern, and
90 * r is the expected result (bit set to 1 if there is difference with the
91 * center, 0 otherwise) */
92 #define P(m, r) ((k_shuffled & (m)) == (r))
93
94 /* adjust 012345678 to 01235678: the mask doesn't contain the (null) diff
95 * between the center/current pixel and itself */
96 #define DROP4(z) ((z) > 4 ? (z)-1 : (z))
97
98 /* shuffle the input mask: move bit n (4-adjusted) to position stored in p<n> */
99 #define SHF(x, rot, n) (((x) >> ((rot) ? 7-DROP4(n) : DROP4(n)) & 1) << DROP4(p##n))
100
101 /* used to check if there is YUV difference between 2 pixels */
102 #define WDIFF(c1, c2) yuv_diff(rgb2yuv(r2y, c1), rgb2yuv(r2y, c2))
103
104 /* bootstrap template for every interpolation code. It defines the shuffled
105 * masks and surrounding pixels. The rot flag is used to indicate if it's a
106 * rotation; its basic effect is to shuffle k using p8..p0 instead of p0..p8 */
107 #define INTERP_BOOTSTRAP(rot) \
108 const int k_shuffled = SHF(k,rot,0) | SHF(k,rot,1) | SHF(k,rot,2) \
109 | SHF(k,rot,3) | 0 | SHF(k,rot,5) \
110 | SHF(k,rot,6) | SHF(k,rot,7) | SHF(k,rot,8); \
111 \
112 const uint32_t w0 = w[p0], w1 = w[p1], \
113 w3 = w[p3], w4 = w[p4], w5 = w[p5], \
114 w7 = w[p7]
115
116 /* Assuming p0..p8 is mapped to pixels 0..8, this function interpolates the
117 * top-left pixel in the total of the 2x2 pixels to interpolates. The function
118 * is also used for the 3 other pixels */
hq2x_interp_1x1(const uint32_t * r2y,int k,const uint32_t * w,int p0,int p1,int p2,int p3,int p4,int p5,int p6,int p7,int p8)119 static av_always_inline uint32_t hq2x_interp_1x1(const uint32_t *r2y, int k,
120 const uint32_t *w,
121 int p0, int p1, int p2,
122 int p3, int p4, int p5,
123 int p6, int p7, int p8)
124 {
125 INTERP_BOOTSTRAP(0);
126
127 if ((P(0xbf,0x37) || P(0xdb,0x13)) && WDIFF(w1, w5))
128 return interp_2px(w4, 3, w3, 1, 2);
129 if ((P(0xdb,0x49) || P(0xef,0x6d)) && WDIFF(w7, w3))
130 return interp_2px(w4, 3, w1, 1, 2);
131 if ((P(0x0b,0x0b) || P(0xfe,0x4a) || P(0xfe,0x1a)) && WDIFF(w3, w1))
132 return w4;
133 if ((P(0x6f,0x2a) || P(0x5b,0x0a) || P(0xbf,0x3a) || P(0xdf,0x5a) ||
134 P(0x9f,0x8a) || P(0xcf,0x8a) || P(0xef,0x4e) || P(0x3f,0x0e) ||
135 P(0xfb,0x5a) || P(0xbb,0x8a) || P(0x7f,0x5a) || P(0xaf,0x8a) ||
136 P(0xeb,0x8a)) && WDIFF(w3, w1))
137 return interp_2px(w4, 3, w0, 1, 2);
138 if (P(0x0b,0x08))
139 return interp_3px(w4, 2, w0, 1, w1, 1, 2);
140 if (P(0x0b,0x02))
141 return interp_3px(w4, 2, w0, 1, w3, 1, 2);
142 if (P(0x2f,0x2f))
143 return interp_3px(w4, 14, w3, 1, w1, 1, 4);
144 if (P(0xbf,0x37) || P(0xdb,0x13))
145 return interp_3px(w4, 5, w1, 2, w3, 1, 3);
146 if (P(0xdb,0x49) || P(0xef,0x6d))
147 return interp_3px(w4, 5, w3, 2, w1, 1, 3);
148 if (P(0x1b,0x03) || P(0x4f,0x43) || P(0x8b,0x83) || P(0x6b,0x43))
149 return interp_2px(w4, 3, w3, 1, 2);
150 if (P(0x4b,0x09) || P(0x8b,0x89) || P(0x1f,0x19) || P(0x3b,0x19))
151 return interp_2px(w4, 3, w1, 1, 2);
152 if (P(0x7e,0x2a) || P(0xef,0xab) || P(0xbf,0x8f) || P(0x7e,0x0e))
153 return interp_3px(w4, 2, w3, 3, w1, 3, 3);
154 if (P(0xfb,0x6a) || P(0x6f,0x6e) || P(0x3f,0x3e) || P(0xfb,0xfa) ||
155 P(0xdf,0xde) || P(0xdf,0x1e))
156 return interp_2px(w4, 3, w0, 1, 2);
157 if (P(0x0a,0x00) || P(0x4f,0x4b) || P(0x9f,0x1b) || P(0x2f,0x0b) ||
158 P(0xbe,0x0a) || P(0xee,0x0a) || P(0x7e,0x0a) || P(0xeb,0x4b) ||
159 P(0x3b,0x1b))
160 return interp_3px(w4, 2, w3, 1, w1, 1, 2);
161 return interp_3px(w4, 6, w3, 1, w1, 1, 3);
162 }
163
164 /* Assuming p0..p8 is mapped to pixels 0..8, this function interpolates the
165 * top-left and top-center pixel in the total of the 3x3 pixels to
166 * interpolates. The function is also used for the 3 other couples of pixels
167 * defining the outline. The center pixel is not defined through this function,
168 * since it's just the same as the original value. */
hq3x_interp_2x1(uint32_t * dst,int dst_linesize,const uint32_t * r2y,int k,const uint32_t * w,int pos00,int pos01,int p0,int p1,int p2,int p3,int p4,int p5,int p6,int p7,int p8,int rotate)169 static av_always_inline void hq3x_interp_2x1(uint32_t *dst, int dst_linesize,
170 const uint32_t *r2y, int k,
171 const uint32_t *w,
172 int pos00, int pos01,
173 int p0, int p1, int p2,
174 int p3, int p4, int p5,
175 int p6, int p7, int p8,
176 int rotate)
177 {
178 INTERP_BOOTSTRAP(rotate);
179
180 uint32_t *dst00 = &dst[dst_linesize*(pos00>>1) + (pos00&1)];
181 uint32_t *dst01 = &dst[dst_linesize*(pos01>>1) + (pos01&1)];
182
183 if ((P(0xdb,0x49) || P(0xef,0x6d)) && WDIFF(w7, w3))
184 *dst00 = interp_2px(w4, 3, w1, 1, 2);
185 else if ((P(0xbf,0x37) || P(0xdb,0x13)) && WDIFF(w1, w5))
186 *dst00 = interp_2px(w4, 3, w3, 1, 2);
187 else if ((P(0x0b,0x0b) || P(0xfe,0x4a) || P(0xfe,0x1a)) && WDIFF(w3, w1))
188 *dst00 = w4;
189 else if ((P(0x6f,0x2a) || P(0x5b,0x0a) || P(0xbf,0x3a) || P(0xdf,0x5a) ||
190 P(0x9f,0x8a) || P(0xcf,0x8a) || P(0xef,0x4e) || P(0x3f,0x0e) ||
191 P(0xfb,0x5a) || P(0xbb,0x8a) || P(0x7f,0x5a) || P(0xaf,0x8a) ||
192 P(0xeb,0x8a)) && WDIFF(w3, w1))
193 *dst00 = interp_2px(w4, 3, w0, 1, 2);
194 else if (P(0x4b,0x09) || P(0x8b,0x89) || P(0x1f,0x19) || P(0x3b,0x19))
195 *dst00 = interp_2px(w4, 3, w1, 1, 2);
196 else if (P(0x1b,0x03) || P(0x4f,0x43) || P(0x8b,0x83) || P(0x6b,0x43))
197 *dst00 = interp_2px(w4, 3, w3, 1, 2);
198 else if (P(0x7e,0x2a) || P(0xef,0xab) || P(0xbf,0x8f) || P(0x7e,0x0e))
199 *dst00 = interp_2px(w3, 1, w1, 1, 1);
200 else if (P(0x4f,0x4b) || P(0x9f,0x1b) || P(0x2f,0x0b) || P(0xbe,0x0a) ||
201 P(0xee,0x0a) || P(0x7e,0x0a) || P(0xeb,0x4b) || P(0x3b,0x1b))
202 *dst00 = interp_3px(w4, 2, w3, 7, w1, 7, 4);
203 else if (P(0x0b,0x08) || P(0xf9,0x68) || P(0xf3,0x62) || P(0x6d,0x6c) ||
204 P(0x67,0x66) || P(0x3d,0x3c) || P(0x37,0x36) || P(0xf9,0xf8) ||
205 P(0xdd,0xdc) || P(0xf3,0xf2) || P(0xd7,0xd6) || P(0xdd,0x1c) ||
206 P(0xd7,0x16) || P(0x0b,0x02))
207 *dst00 = interp_2px(w4, 3, w0, 1, 2);
208 else
209 *dst00 = interp_3px(w4, 2, w3, 1, w1, 1, 2);
210
211 if ((P(0xfe,0xde) || P(0x9e,0x16) || P(0xda,0x12) || P(0x17,0x16) ||
212 P(0x5b,0x12) || P(0xbb,0x12)) && WDIFF(w1, w5))
213 *dst01 = w4;
214 else if ((P(0x0f,0x0b) || P(0x5e,0x0a) || P(0xfb,0x7b) || P(0x3b,0x0b) ||
215 P(0xbe,0x0a) || P(0x7a,0x0a)) && WDIFF(w3, w1))
216 *dst01 = w4;
217 else if (P(0xbf,0x8f) || P(0x7e,0x0e) || P(0xbf,0x37) || P(0xdb,0x13))
218 *dst01 = interp_2px(w1, 3, w4, 1, 2);
219 else if (P(0x02,0x00) || P(0x7c,0x28) || P(0xed,0xa9) || P(0xf5,0xb4) ||
220 P(0xd9,0x90))
221 *dst01 = interp_2px(w4, 3, w1, 1, 2);
222 else if (P(0x4f,0x4b) || P(0xfb,0x7b) || P(0xfe,0x7e) || P(0x9f,0x1b) ||
223 P(0x2f,0x0b) || P(0xbe,0x0a) || P(0x7e,0x0a) || P(0xfb,0x4b) ||
224 P(0xfb,0xdb) || P(0xfe,0xde) || P(0xfe,0x56) || P(0x57,0x56) ||
225 P(0x97,0x16) || P(0x3f,0x1e) || P(0xdb,0x12) || P(0xbb,0x12))
226 *dst01 = interp_2px(w4, 7, w1, 1, 3);
227 else
228 *dst01 = w4;
229 }
230
231 /* Assuming p0..p8 is mapped to pixels 0..8, this function interpolates the
232 * top-left block of 2x2 pixels in the total of the 4x4 pixels (or 4 blocks) to
233 * interpolates. The function is also used for the 3 other blocks of 2x2
234 * pixels. */
hq4x_interp_2x2(uint32_t * dst,int dst_linesize,const uint32_t * r2y,int k,const uint32_t * w,int pos00,int pos01,int pos10,int pos11,int p0,int p1,int p2,int p3,int p4,int p5,int p6,int p7,int p8)235 static av_always_inline void hq4x_interp_2x2(uint32_t *dst, int dst_linesize,
236 const uint32_t *r2y, int k,
237 const uint32_t *w,
238 int pos00, int pos01,
239 int pos10, int pos11,
240 int p0, int p1, int p2,
241 int p3, int p4, int p5,
242 int p6, int p7, int p8)
243 {
244 INTERP_BOOTSTRAP(0);
245
246 uint32_t *dst00 = &dst[dst_linesize*(pos00>>1) + (pos00&1)];
247 uint32_t *dst01 = &dst[dst_linesize*(pos01>>1) + (pos01&1)];
248 uint32_t *dst10 = &dst[dst_linesize*(pos10>>1) + (pos10&1)];
249 uint32_t *dst11 = &dst[dst_linesize*(pos11>>1) + (pos11&1)];
250
251 const int cond00 = (P(0xbf,0x37) || P(0xdb,0x13)) && WDIFF(w1, w5);
252 const int cond01 = (P(0xdb,0x49) || P(0xef,0x6d)) && WDIFF(w7, w3);
253 const int cond02 = (P(0x6f,0x2a) || P(0x5b,0x0a) || P(0xbf,0x3a) ||
254 P(0xdf,0x5a) || P(0x9f,0x8a) || P(0xcf,0x8a) ||
255 P(0xef,0x4e) || P(0x3f,0x0e) || P(0xfb,0x5a) ||
256 P(0xbb,0x8a) || P(0x7f,0x5a) || P(0xaf,0x8a) ||
257 P(0xeb,0x8a)) && WDIFF(w3, w1);
258 const int cond03 = P(0xdb,0x49) || P(0xef,0x6d);
259 const int cond04 = P(0xbf,0x37) || P(0xdb,0x13);
260 const int cond05 = P(0x1b,0x03) || P(0x4f,0x43) || P(0x8b,0x83) ||
261 P(0x6b,0x43);
262 const int cond06 = P(0x4b,0x09) || P(0x8b,0x89) || P(0x1f,0x19) ||
263 P(0x3b,0x19);
264 const int cond07 = P(0x0b,0x08) || P(0xf9,0x68) || P(0xf3,0x62) ||
265 P(0x6d,0x6c) || P(0x67,0x66) || P(0x3d,0x3c) ||
266 P(0x37,0x36) || P(0xf9,0xf8) || P(0xdd,0xdc) ||
267 P(0xf3,0xf2) || P(0xd7,0xd6) || P(0xdd,0x1c) ||
268 P(0xd7,0x16) || P(0x0b,0x02);
269 const int cond08 = (P(0x0f,0x0b) || P(0x2b,0x0b) || P(0xfe,0x4a) ||
270 P(0xfe,0x1a)) && WDIFF(w3, w1);
271 const int cond09 = P(0x2f,0x2f);
272 const int cond10 = P(0x0a,0x00);
273 const int cond11 = P(0x0b,0x09);
274 const int cond12 = P(0x7e,0x2a) || P(0xef,0xab);
275 const int cond13 = P(0xbf,0x8f) || P(0x7e,0x0e);
276 const int cond14 = P(0x4f,0x4b) || P(0x9f,0x1b) || P(0x2f,0x0b) ||
277 P(0xbe,0x0a) || P(0xee,0x0a) || P(0x7e,0x0a) ||
278 P(0xeb,0x4b) || P(0x3b,0x1b);
279 const int cond15 = P(0x0b,0x03);
280
281 if (cond00)
282 *dst00 = interp_2px(w4, 5, w3, 3, 3);
283 else if (cond01)
284 *dst00 = interp_2px(w4, 5, w1, 3, 3);
285 else if ((P(0x0b,0x0b) || P(0xfe,0x4a) || P(0xfe,0x1a)) && WDIFF(w3, w1))
286 *dst00 = w4;
287 else if (cond02)
288 *dst00 = interp_2px(w4, 5, w0, 3, 3);
289 else if (cond03)
290 *dst00 = interp_2px(w4, 3, w3, 1, 2);
291 else if (cond04)
292 *dst00 = interp_2px(w4, 3, w1, 1, 2);
293 else if (cond05)
294 *dst00 = interp_2px(w4, 5, w3, 3, 3);
295 else if (cond06)
296 *dst00 = interp_2px(w4, 5, w1, 3, 3);
297 else if (P(0x0f,0x0b) || P(0x5e,0x0a) || P(0x2b,0x0b) || P(0xbe,0x0a) ||
298 P(0x7a,0x0a) || P(0xee,0x0a))
299 *dst00 = interp_2px(w1, 1, w3, 1, 1);
300 else if (cond07)
301 *dst00 = interp_2px(w4, 5, w0, 3, 3);
302 else
303 *dst00 = interp_3px(w4, 2, w1, 1, w3, 1, 2);
304
305 if (cond00)
306 *dst01 = interp_2px(w4, 7, w3, 1, 3);
307 else if (cond08)
308 *dst01 = w4;
309 else if (cond02)
310 *dst01 = interp_2px(w4, 3, w0, 1, 2);
311 else if (cond09)
312 *dst01 = w4;
313 else if (cond10)
314 *dst01 = interp_3px(w4, 5, w1, 2, w3, 1, 3);
315 else if (P(0x0b,0x08))
316 *dst01 = interp_3px(w4, 5, w1, 2, w0, 1, 3);
317 else if (cond11)
318 *dst01 = interp_2px(w4, 5, w1, 3, 3);
319 else if (cond04)
320 *dst01 = interp_2px(w1, 3, w4, 1, 2);
321 else if (cond12)
322 *dst01 = interp_3px(w1, 2, w4, 1, w3, 1, 2);
323 else if (cond13)
324 *dst01 = interp_2px(w1, 5, w3, 3, 3);
325 else if (cond05)
326 *dst01 = interp_2px(w4, 7, w3, 1, 3);
327 else if (P(0xf3,0x62) || P(0x67,0x66) || P(0x37,0x36) || P(0xf3,0xf2) ||
328 P(0xd7,0xd6) || P(0xd7,0x16) || P(0x0b,0x02))
329 *dst01 = interp_2px(w4, 3, w0, 1, 2);
330 else if (cond14)
331 *dst01 = interp_2px(w1, 1, w4, 1, 1);
332 else
333 *dst01 = interp_2px(w4, 3, w1, 1, 2);
334
335 if (cond01)
336 *dst10 = interp_2px(w4, 7, w1, 1, 3);
337 else if (cond08)
338 *dst10 = w4;
339 else if (cond02)
340 *dst10 = interp_2px(w4, 3, w0, 1, 2);
341 else if (cond09)
342 *dst10 = w4;
343 else if (cond10)
344 *dst10 = interp_3px(w4, 5, w3, 2, w1, 1, 3);
345 else if (P(0x0b,0x02))
346 *dst10 = interp_3px(w4, 5, w3, 2, w0, 1, 3);
347 else if (cond15)
348 *dst10 = interp_2px(w4, 5, w3, 3, 3);
349 else if (cond03)
350 *dst10 = interp_2px(w3, 3, w4, 1, 2);
351 else if (cond13)
352 *dst10 = interp_3px(w3, 2, w4, 1, w1, 1, 2);
353 else if (cond12)
354 *dst10 = interp_2px(w3, 5, w1, 3, 3);
355 else if (cond06)
356 *dst10 = interp_2px(w4, 7, w1, 1, 3);
357 else if (P(0x0b,0x08) || P(0xf9,0x68) || P(0x6d,0x6c) || P(0x3d,0x3c) ||
358 P(0xf9,0xf8) || P(0xdd,0xdc) || P(0xdd,0x1c))
359 *dst10 = interp_2px(w4, 3, w0, 1, 2);
360 else if (cond14)
361 *dst10 = interp_2px(w3, 1, w4, 1, 1);
362 else
363 *dst10 = interp_2px(w4, 3, w3, 1, 2);
364
365 if ((P(0x7f,0x2b) || P(0xef,0xab) || P(0xbf,0x8f) || P(0x7f,0x0f)) &&
366 WDIFF(w3, w1))
367 *dst11 = w4;
368 else if (cond02)
369 *dst11 = interp_2px(w4, 7, w0, 1, 3);
370 else if (cond15)
371 *dst11 = interp_2px(w4, 7, w3, 1, 3);
372 else if (cond11)
373 *dst11 = interp_2px(w4, 7, w1, 1, 3);
374 else if (P(0x0a,0x00) || P(0x7e,0x2a) || P(0xef,0xab) || P(0xbf,0x8f) ||
375 P(0x7e,0x0e))
376 *dst11 = interp_3px(w4, 6, w3, 1, w1, 1, 3);
377 else if (cond07)
378 *dst11 = interp_2px(w4, 7, w0, 1, 3);
379 else
380 *dst11 = w4;
381 }
382
hqx_filter(const ThreadData * td,int jobnr,int nb_jobs,int n)383 static av_always_inline void hqx_filter(const ThreadData *td, int jobnr, int nb_jobs, int n)
384 {
385 int x, y;
386 AVFrame *in = td->in, *out = td->out;
387 const uint32_t *r2y = td->rgbtoyuv;
388 const int height = in->height;
389 const int width = in->width;
390 const int slice_start = (height * jobnr ) / nb_jobs;
391 const int slice_end = (height * (jobnr+1)) / nb_jobs;
392 const int dst_linesize = out->linesize[0];
393 const int src_linesize = in->linesize[0];
394 uint8_t *dst = out->data[0] + slice_start * dst_linesize * n;
395 const uint8_t *src = in->data[0] + slice_start * src_linesize;
396
397 const int dst32_linesize = dst_linesize >> 2;
398 const int src32_linesize = src_linesize >> 2;
399
400 for (y = slice_start; y < slice_end; y++) {
401 const uint32_t *src32 = (const uint32_t *)src;
402 uint32_t *dst32 = (uint32_t *)dst;
403 const int prevline = y > 0 ? -src32_linesize : 0;
404 const int nextline = y < height - 1 ? src32_linesize : 0;
405
406 for (x = 0; x < width; x++) {
407 const int prevcol = x > 0 ? -1 : 0;
408 const int nextcol = x < width -1 ? 1 : 0;
409 const uint32_t w[3*3] = {
410 src32[prevcol + prevline], src32[prevline], src32[prevline + nextcol],
411 src32[prevcol ], src32[ 0], src32[ nextcol],
412 src32[prevcol + nextline], src32[nextline], src32[nextline + nextcol]
413 };
414 const uint32_t yuv1 = rgb2yuv(r2y, w[4]);
415 const int pattern = (w[4] != w[0] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[0]))) : 0)
416 | (w[4] != w[1] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[1]))) : 0) << 1
417 | (w[4] != w[2] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[2]))) : 0) << 2
418 | (w[4] != w[3] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[3]))) : 0) << 3
419 | (w[4] != w[5] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[5]))) : 0) << 4
420 | (w[4] != w[6] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[6]))) : 0) << 5
421 | (w[4] != w[7] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[7]))) : 0) << 6
422 | (w[4] != w[8] ? (yuv_diff(yuv1, rgb2yuv(r2y, w[8]))) : 0) << 7;
423
424 if (n == 2) {
425 dst32[dst32_linesize*0 + 0] = hq2x_interp_1x1(r2y, pattern, w, 0,1,2,3,4,5,6,7,8); // 00
426 dst32[dst32_linesize*0 + 1] = hq2x_interp_1x1(r2y, pattern, w, 2,1,0,5,4,3,8,7,6); // 01 (vert mirrored)
427 dst32[dst32_linesize*1 + 0] = hq2x_interp_1x1(r2y, pattern, w, 6,7,8,3,4,5,0,1,2); // 10 (horiz mirrored)
428 dst32[dst32_linesize*1 + 1] = hq2x_interp_1x1(r2y, pattern, w, 8,7,6,5,4,3,2,1,0); // 11 (center mirrored)
429 } else if (n == 3) {
430 hq3x_interp_2x1(dst32, dst32_linesize, r2y, pattern, w, 0,1, 0,1,2,3,4,5,6,7,8, 0); // 00 01
431 hq3x_interp_2x1(dst32 + 1, dst32_linesize, r2y, pattern, w, 1,3, 2,5,8,1,4,7,0,3,6, 1); // 02 12 (rotated to the right)
432 hq3x_interp_2x1(dst32 + 1*dst32_linesize, dst32_linesize, r2y, pattern, w, 2,0, 6,3,0,7,4,1,8,5,2, 1); // 20 10 (rotated to the left)
433 hq3x_interp_2x1(dst32 + 1*dst32_linesize + 1, dst32_linesize, r2y, pattern, w, 3,2, 8,7,6,5,4,3,2,1,0, 0); // 22 21 (center mirrored)
434 dst32[dst32_linesize + 1] = w[4]; // 11
435 } else if (n == 4) {
436 hq4x_interp_2x2(dst32, dst32_linesize, r2y, pattern, w, 0,1,2,3, 0,1,2,3,4,5,6,7,8); // 00 01 10 11
437 hq4x_interp_2x2(dst32 + 2, dst32_linesize, r2y, pattern, w, 1,0,3,2, 2,1,0,5,4,3,8,7,6); // 02 03 12 13 (vert mirrored)
438 hq4x_interp_2x2(dst32 + 2*dst32_linesize, dst32_linesize, r2y, pattern, w, 2,3,0,1, 6,7,8,3,4,5,0,1,2); // 20 21 30 31 (horiz mirrored)
439 hq4x_interp_2x2(dst32 + 2*dst32_linesize + 2, dst32_linesize, r2y, pattern, w, 3,2,1,0, 8,7,6,5,4,3,2,1,0); // 22 23 32 33 (center mirrored)
440 } else {
441 av_assert0(0);
442 }
443
444 src32 += 1;
445 dst32 += n;
446 }
447
448 src += src_linesize;
449 dst += dst_linesize * n;
450 }
451 }
452
453 #define HQX_FUNC(size) \
454 static int hq##size##x(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
455 { \
456 hqx_filter(arg, jobnr, nb_jobs, size); \
457 return 0; \
458 }
459
460 HQX_FUNC(2)
461 HQX_FUNC(3)
462 HQX_FUNC(4)
463
config_output(AVFilterLink * outlink)464 static int config_output(AVFilterLink *outlink)
465 {
466 AVFilterContext *ctx = outlink->src;
467 HQXContext *hqx = ctx->priv;
468 AVFilterLink *inlink = ctx->inputs[0];
469
470 outlink->w = inlink->w * hqx->n;
471 outlink->h = inlink->h * hqx->n;
472 av_log(inlink->dst, AV_LOG_VERBOSE, "fmt:%s size:%dx%d -> size:%dx%d\n",
473 av_get_pix_fmt_name(inlink->format),
474 inlink->w, inlink->h, outlink->w, outlink->h);
475 return 0;
476 }
477
filter_frame(AVFilterLink * inlink,AVFrame * in)478 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
479 {
480 AVFilterContext *ctx = inlink->dst;
481 AVFilterLink *outlink = ctx->outputs[0];
482 HQXContext *hqx = ctx->priv;
483 ThreadData td;
484 AVFrame *out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
485 if (!out) {
486 av_frame_free(&in);
487 return AVERROR(ENOMEM);
488 }
489 av_frame_copy_props(out, in);
490 out->width = outlink->w;
491 out->height = outlink->h;
492
493 td.in = in;
494 td.out = out;
495 td.rgbtoyuv = hqx->rgbtoyuv;
496 ff_filter_execute(ctx, hqx->func, &td, NULL,
497 FFMIN(inlink->h, ff_filter_get_nb_threads(ctx)));
498
499 av_frame_free(&in);
500 return ff_filter_frame(outlink, out);
501 }
502
init(AVFilterContext * ctx)503 static av_cold int init(AVFilterContext *ctx)
504 {
505 HQXContext *hqx = ctx->priv;
506 static const hqxfunc_t hqxfuncs[] = {hq2x, hq3x, hq4x};
507
508 uint32_t c;
509 int bg, rg, g;
510
511 for (bg=-255; bg<256; bg++) {
512 for (rg=-255; rg<256; rg++) {
513 const uint32_t u = (uint32_t)((-169*rg + 500*bg)/1000) + 128;
514 const uint32_t v = (uint32_t)(( 500*rg - 81*bg)/1000) + 128;
515 int startg = FFMAX3(-bg, -rg, 0);
516 int endg = FFMIN3(255-bg, 255-rg, 255);
517 uint32_t y = (uint32_t)(( 299*rg + 1000*startg + 114*bg)/1000);
518 c = bg + rg * (1 << 16) + 0x010101 * startg;
519 for (g = startg; g <= endg; g++) {
520 hqx->rgbtoyuv[c] = ((y++) << 16) + (u << 8) + v;
521 c+= 0x010101;
522 }
523 }
524 }
525
526 hqx->func = hqxfuncs[hqx->n - 2];
527 return 0;
528 }
529
530 static const AVFilterPad hqx_inputs[] = {
531 {
532 .name = "default",
533 .type = AVMEDIA_TYPE_VIDEO,
534 .filter_frame = filter_frame,
535 },
536 };
537
538 static const AVFilterPad hqx_outputs[] = {
539 {
540 .name = "default",
541 .type = AVMEDIA_TYPE_VIDEO,
542 .config_props = config_output,
543 },
544 };
545
546 const AVFilter ff_vf_hqx = {
547 .name = "hqx",
548 .description = NULL_IF_CONFIG_SMALL("Scale the input by 2, 3 or 4 using the hq*x magnification algorithm."),
549 .priv_size = sizeof(HQXContext),
550 .init = init,
551 FILTER_INPUTS(hqx_inputs),
552 FILTER_OUTPUTS(hqx_outputs),
553 FILTER_SINGLE_PIXFMT(AV_PIX_FMT_RGB32),
554 .priv_class = &hqx_class,
555 .flags = AVFILTER_FLAG_SLICE_THREADS,
556 };
557