30421
|
1 /*
|
|
2 * This file is part of MPlayer.
|
|
3 *
|
|
4 * MPlayer is free software; you can redistribute it and/or modify
|
|
5 * it under the terms of the GNU General Public License as published by
|
|
6 * the Free Software Foundation; either version 2 of the License, or
|
|
7 * (at your option) any later version.
|
|
8 *
|
|
9 * MPlayer is distributed in the hope that it will be useful,
|
|
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
12 * GNU General Public License for more details.
|
|
13 *
|
|
14 * You should have received a copy of the GNU General Public License along
|
|
15 * with MPlayer; if not, write to the Free Software Foundation, Inc.,
|
|
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
|
17 */
|
10664
|
18
|
11412
|
19 #include <stdio.h>
|
10664
|
20 #include <stdlib.h>
|
11412
|
21 #include <string.h>
|
26203
|
22 #include "config.h"
|
10664
|
23 #include "pullup.h"
|
20961
|
24 #include "cpudetect.h"
|
10664
|
25
|
|
26
|
|
27
|
28290
|
28 #if ARCH_X86
|
|
29 #if HAVE_MMX
|
10664
|
30 static int diff_y_mmx(unsigned char *a, unsigned char *b, int s)
|
|
31 {
|
32702
|
32 int ret;
|
|
33 __asm__ volatile (
|
|
34 "movl $4, %%ecx \n\t"
|
|
35 "pxor %%mm4, %%mm4 \n\t"
|
|
36 "pxor %%mm7, %%mm7 \n\t"
|
29263
|
37
|
32702
|
38 "1: \n\t"
|
29263
|
39
|
32702
|
40 "movq (%%"REG_S"), %%mm0 \n\t"
|
|
41 "movq (%%"REG_S"), %%mm2 \n\t"
|
|
42 "add %%"REG_a", %%"REG_S" \n\t"
|
|
43 "movq (%%"REG_D"), %%mm1 \n\t"
|
|
44 "add %%"REG_a", %%"REG_D" \n\t"
|
|
45 "psubusb %%mm1, %%mm2 \n\t"
|
|
46 "psubusb %%mm0, %%mm1 \n\t"
|
|
47 "movq %%mm2, %%mm0 \n\t"
|
|
48 "movq %%mm1, %%mm3 \n\t"
|
|
49 "punpcklbw %%mm7, %%mm0 \n\t"
|
|
50 "punpcklbw %%mm7, %%mm1 \n\t"
|
|
51 "punpckhbw %%mm7, %%mm2 \n\t"
|
|
52 "punpckhbw %%mm7, %%mm3 \n\t"
|
|
53 "paddw %%mm0, %%mm4 \n\t"
|
|
54 "paddw %%mm1, %%mm4 \n\t"
|
|
55 "paddw %%mm2, %%mm4 \n\t"
|
|
56 "paddw %%mm3, %%mm4 \n\t"
|
29263
|
57
|
32702
|
58 "decl %%ecx \n\t"
|
|
59 "jnz 1b \n\t"
|
10741
|
60
|
32702
|
61 "movq %%mm4, %%mm3 \n\t"
|
|
62 "punpcklwd %%mm7, %%mm4 \n\t"
|
|
63 "punpckhwd %%mm7, %%mm3 \n\t"
|
|
64 "paddd %%mm4, %%mm3 \n\t"
|
|
65 "movd %%mm3, %%eax \n\t"
|
|
66 "psrlq $32, %%mm3 \n\t"
|
|
67 "movd %%mm3, %%edx \n\t"
|
|
68 "addl %%edx, %%eax \n\t"
|
|
69 "emms \n\t"
|
|
70 : "=a" (ret)
|
|
71 : "S" (a), "D" (b), "a" (s)
|
|
72 : "%ecx", "%edx"
|
|
73 );
|
|
74 return ret;
|
10747
|
75 }
|
|
76
|
|
77 static int licomb_y_mmx(unsigned char *a, unsigned char *b, int s)
|
|
78 {
|
32702
|
79 int ret;
|
|
80 __asm__ volatile (
|
|
81 "movl $4, %%ecx \n\t"
|
|
82 "pxor %%mm6, %%mm6 \n\t"
|
|
83 "pxor %%mm7, %%mm7 \n\t"
|
|
84 "sub %%"REG_a", %%"REG_D" \n\t"
|
29263
|
85
|
32702
|
86 "2: \n\t"
|
10747
|
87
|
32702
|
88 "movq (%%"REG_D"), %%mm0 \n\t"
|
|
89 "movq (%%"REG_D"), %%mm1 \n\t"
|
|
90 "punpcklbw %%mm7, %%mm0 \n\t"
|
|
91 "movq (%%"REG_D",%%"REG_a"), %%mm2 \n\t"
|
|
92 "punpcklbw %%mm7, %%mm1 \n\t"
|
|
93 "punpcklbw %%mm7, %%mm2 \n\t"
|
|
94 "paddw %%mm0, %%mm0 \n\t"
|
|
95 "paddw %%mm2, %%mm1 \n\t"
|
|
96 "movq %%mm0, %%mm2 \n\t"
|
|
97 "psubusw %%mm1, %%mm0 \n\t"
|
|
98 "psubusw %%mm2, %%mm1 \n\t"
|
|
99 "paddw %%mm0, %%mm6 \n\t"
|
|
100 "paddw %%mm1, %%mm6 \n\t"
|
10747
|
101
|
32702
|
102 "movq (%%"REG_S"), %%mm0 \n\t"
|
|
103 "movq (%%"REG_D"), %%mm1 \n\t"
|
|
104 "punpckhbw %%mm7, %%mm0 \n\t"
|
|
105 "movq (%%"REG_D",%%"REG_a"), %%mm2 \n\t"
|
|
106 "punpckhbw %%mm7, %%mm1 \n\t"
|
|
107 "punpckhbw %%mm7, %%mm2 \n\t"
|
|
108 "paddw %%mm0, %%mm0 \n\t"
|
|
109 "paddw %%mm2, %%mm1 \n\t"
|
|
110 "movq %%mm0, %%mm2 \n\t"
|
|
111 "psubusw %%mm1, %%mm0 \n\t"
|
|
112 "psubusw %%mm2, %%mm1 \n\t"
|
|
113 "paddw %%mm0, %%mm6 \n\t"
|
|
114 "paddw %%mm1, %%mm6 \n\t"
|
29263
|
115
|
32702
|
116 "movq (%%"REG_D",%%"REG_a"), %%mm0 \n\t"
|
|
117 "movq (%%"REG_S"), %%mm1 \n\t"
|
|
118 "punpcklbw %%mm7, %%mm0 \n\t"
|
|
119 "movq (%%"REG_S",%%"REG_a"), %%mm2 \n\t"
|
|
120 "punpcklbw %%mm7, %%mm1 \n\t"
|
|
121 "punpcklbw %%mm7, %%mm2 \n\t"
|
|
122 "paddw %%mm0, %%mm0 \n\t"
|
|
123 "paddw %%mm2, %%mm1 \n\t"
|
|
124 "movq %%mm0, %%mm2 \n\t"
|
|
125 "psubusw %%mm1, %%mm0 \n\t"
|
|
126 "psubusw %%mm2, %%mm1 \n\t"
|
|
127 "paddw %%mm0, %%mm6 \n\t"
|
|
128 "paddw %%mm1, %%mm6 \n\t"
|
29263
|
129
|
32702
|
130 "movq (%%"REG_D",%%"REG_a"), %%mm0 \n\t"
|
|
131 "movq (%%"REG_S"), %%mm1 \n\t"
|
|
132 "punpckhbw %%mm7, %%mm0 \n\t"
|
|
133 "movq (%%"REG_S",%%"REG_a"), %%mm2 \n\t"
|
|
134 "punpckhbw %%mm7, %%mm1 \n\t"
|
|
135 "punpckhbw %%mm7, %%mm2 \n\t"
|
|
136 "paddw %%mm0, %%mm0 \n\t"
|
|
137 "paddw %%mm2, %%mm1 \n\t"
|
|
138 "movq %%mm0, %%mm2 \n\t"
|
|
139 "psubusw %%mm1, %%mm0 \n\t"
|
|
140 "psubusw %%mm2, %%mm1 \n\t"
|
|
141 "paddw %%mm0, %%mm6 \n\t"
|
|
142 "paddw %%mm1, %%mm6 \n\t"
|
10747
|
143
|
32702
|
144 "add %%"REG_a", %%"REG_S" \n\t"
|
|
145 "add %%"REG_a", %%"REG_D" \n\t"
|
|
146 "decl %%ecx \n\t"
|
|
147 "jnz 2b \n\t"
|
29263
|
148
|
32702
|
149 "movq %%mm6, %%mm5 \n\t"
|
|
150 "punpcklwd %%mm7, %%mm6 \n\t"
|
|
151 "punpckhwd %%mm7, %%mm5 \n\t"
|
|
152 "paddd %%mm6, %%mm5 \n\t"
|
|
153 "movd %%mm5, %%eax \n\t"
|
|
154 "psrlq $32, %%mm5 \n\t"
|
|
155 "movd %%mm5, %%edx \n\t"
|
|
156 "addl %%edx, %%eax \n\t"
|
29263
|
157
|
32702
|
158 "emms \n\t"
|
|
159 : "=a" (ret)
|
|
160 : "S" (a), "D" (b), "a" (s)
|
|
161 : "%ecx", "%edx"
|
|
162 );
|
|
163 return ret;
|
10664
|
164 }
|
15721
|
165
|
|
166 static int var_y_mmx(unsigned char *a, unsigned char *b, int s)
|
|
167 {
|
32702
|
168 int ret;
|
|
169 __asm__ volatile (
|
|
170 "movl $3, %%ecx \n\t"
|
|
171 "pxor %%mm4, %%mm4 \n\t"
|
|
172 "pxor %%mm7, %%mm7 \n\t"
|
29263
|
173
|
32702
|
174 "1: \n\t"
|
29263
|
175
|
32702
|
176 "movq (%%"REG_S"), %%mm0 \n\t"
|
|
177 "movq (%%"REG_S"), %%mm2 \n\t"
|
|
178 "movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
|
|
179 "add %%"REG_a", %%"REG_S" \n\t"
|
|
180 "psubusb %%mm1, %%mm2 \n\t"
|
|
181 "psubusb %%mm0, %%mm1 \n\t"
|
|
182 "movq %%mm2, %%mm0 \n\t"
|
|
183 "movq %%mm1, %%mm3 \n\t"
|
|
184 "punpcklbw %%mm7, %%mm0 \n\t"
|
|
185 "punpcklbw %%mm7, %%mm1 \n\t"
|
|
186 "punpckhbw %%mm7, %%mm2 \n\t"
|
|
187 "punpckhbw %%mm7, %%mm3 \n\t"
|
|
188 "paddw %%mm0, %%mm4 \n\t"
|
|
189 "paddw %%mm1, %%mm4 \n\t"
|
|
190 "paddw %%mm2, %%mm4 \n\t"
|
|
191 "paddw %%mm3, %%mm4 \n\t"
|
29263
|
192
|
32702
|
193 "decl %%ecx \n\t"
|
|
194 "jnz 1b \n\t"
|
15721
|
195
|
32702
|
196 "movq %%mm4, %%mm3 \n\t"
|
|
197 "punpcklwd %%mm7, %%mm4 \n\t"
|
|
198 "punpckhwd %%mm7, %%mm3 \n\t"
|
|
199 "paddd %%mm4, %%mm3 \n\t"
|
|
200 "movd %%mm3, %%eax \n\t"
|
|
201 "psrlq $32, %%mm3 \n\t"
|
|
202 "movd %%mm3, %%edx \n\t"
|
|
203 "addl %%edx, %%eax \n\t"
|
|
204 "emms \n\t"
|
|
205 : "=a" (ret)
|
|
206 : "S" (a), "a" (s)
|
|
207 : "%ecx", "%edx"
|
|
208 );
|
|
209 return 4*ret;
|
15721
|
210 }
|
10664
|
211 #endif
|
13720
|
212 #endif
|
10664
|
213
|
|
214 #define ABS(a) (((a)^((a)>>31))-((a)>>31))
|
|
215
|
|
216 static int diff_y(unsigned char *a, unsigned char *b, int s)
|
|
217 {
|
32702
|
218 int i, j, diff=0;
|
|
219 for (i=4; i; i--) {
|
|
220 for (j=0; j<8; j++) diff += ABS(a[j]-b[j]);
|
|
221 a+=s; b+=s;
|
|
222 }
|
|
223 return diff;
|
10664
|
224 }
|
|
225
|
|
226 static int licomb_y(unsigned char *a, unsigned char *b, int s)
|
|
227 {
|
32702
|
228 int i, j, diff=0;
|
|
229 for (i=4; i; i--) {
|
|
230 for (j=0; j<8; j++)
|
|
231 diff += ABS((a[j]<<1) - b[j-s] - b[j])
|
|
232 + ABS((b[j]<<1) - a[j] - a[j+s]);
|
|
233 a+=s; b+=s;
|
|
234 }
|
|
235 return diff;
|
10664
|
236 }
|
|
237
|
24319
|
238 #if 0
|
11626
|
239 static int qpcomb_y(unsigned char *a, unsigned char *b, int s)
|
|
240 {
|
32702
|
241 int i, j, diff=0;
|
|
242 for (i=4; i; i--) {
|
|
243 for (j=0; j<8; j++)
|
|
244 diff += ABS(a[j] - 3*b[j-s] + 3*a[j+s] - b[j]);
|
|
245 a+=s; b+=s;
|
|
246 }
|
|
247 return diff;
|
11626
|
248 }
|
|
249
|
10747
|
250 static int licomb_y_test(unsigned char *a, unsigned char *b, int s)
|
|
251 {
|
32702
|
252 int c = licomb_y(a,b,s);
|
|
253 int m = licomb_y_mmx(a,b,s);
|
|
254 if (c != m) printf("%d != %d\n", c, m);
|
|
255 return m;
|
10747
|
256 }
|
10843
|
257 #endif
|
10747
|
258
|
15721
|
259 static int var_y(unsigned char *a, unsigned char *b, int s)
|
|
260 {
|
32702
|
261 int i, j, var=0;
|
|
262 for (i=3; i; i--) {
|
|
263 for (j=0; j<8; j++) {
|
|
264 var += ABS(a[j]-a[j+s]);
|
|
265 }
|
|
266 a+=s; b+=s;
|
|
267 }
|
|
268 return 4*var; /* match comb scaling */
|
15721
|
269 }
|
|
270
|
10664
|
271
|
|
272
|
|
273
|
|
274
|
|
275
|
|
276
|
|
277
|
|
278
|
|
279 static void alloc_buffer(struct pullup_context *c, struct pullup_buffer *b)
|
|
280 {
|
32702
|
281 int i;
|
|
282 if (b->planes) return;
|
|
283 b->planes = calloc(c->nplanes, sizeof(unsigned char *));
|
|
284 for (i = 0; i < c->nplanes; i++) {
|
|
285 b->planes[i] = malloc(c->h[i]*c->stride[i]);
|
|
286 /* Deal with idiotic 128=0 for chroma: */
|
|
287 memset(b->planes[i], c->background[i], c->h[i]*c->stride[i]);
|
|
288 }
|
10664
|
289 }
|
|
290
|
|
291 struct pullup_buffer *pullup_lock_buffer(struct pullup_buffer *b, int parity)
|
|
292 {
|
32702
|
293 if (!b) return 0;
|
|
294 if ((parity+1) & 1) b->lock[0]++;
|
|
295 if ((parity+1) & 2) b->lock[1]++;
|
|
296 return b;
|
10664
|
297 }
|
|
298
|
|
299 void pullup_release_buffer(struct pullup_buffer *b, int parity)
|
|
300 {
|
32702
|
301 if (!b) return;
|
|
302 if ((parity+1) & 1) b->lock[0]--;
|
|
303 if ((parity+1) & 2) b->lock[1]--;
|
10664
|
304 }
|
|
305
|
|
306 struct pullup_buffer *pullup_get_buffer(struct pullup_context *c, int parity)
|
|
307 {
|
32702
|
308 int i;
|
10664
|
309
|
32702
|
310 /* Try first to get the sister buffer for the previous field */
|
|
311 if (parity < 2 && c->last && parity != c->last->parity
|
|
312 && !c->last->buffer->lock[parity]) {
|
|
313 alloc_buffer(c, c->last->buffer);
|
|
314 return pullup_lock_buffer(c->last->buffer, parity);
|
|
315 }
|
29263
|
316
|
32702
|
317 /* Prefer a buffer with both fields open */
|
|
318 for (i = 0; i < c->nbuffers; i++) {
|
|
319 if (c->buffers[i].lock[0]) continue;
|
|
320 if (c->buffers[i].lock[1]) continue;
|
|
321 alloc_buffer(c, &c->buffers[i]);
|
|
322 return pullup_lock_buffer(&c->buffers[i], parity);
|
|
323 }
|
10664
|
324
|
32702
|
325 if (parity == 2) return 0;
|
29263
|
326
|
32702
|
327 /* Search for any half-free buffer */
|
|
328 for (i = 0; i < c->nbuffers; i++) {
|
|
329 if (((parity+1) & 1) && c->buffers[i].lock[0]) continue;
|
|
330 if (((parity+1) & 2) && c->buffers[i].lock[1]) continue;
|
|
331 alloc_buffer(c, &c->buffers[i]);
|
|
332 return pullup_lock_buffer(&c->buffers[i], parity);
|
|
333 }
|
29263
|
334
|
32702
|
335 return 0;
|
10664
|
336 }
|
|
337
|
|
338
|
|
339
|
|
340
|
|
341
|
|
342
|
|
343 static void compute_metric(struct pullup_context *c,
|
32702
|
344 struct pullup_field *fa, int pa,
|
|
345 struct pullup_field *fb, int pb,
|
|
346 int (*func)(unsigned char *, unsigned char *, int), int *dest)
|
10664
|
347 {
|
32702
|
348 unsigned char *a, *b;
|
|
349 int x, y;
|
|
350 int mp = c->metric_plane;
|
|
351 int xstep = c->bpp[mp];
|
|
352 int ystep = c->stride[mp]<<3;
|
|
353 int s = c->stride[mp]<<1; /* field stride */
|
|
354 int w = c->metric_w*xstep;
|
10664
|
355
|
32702
|
356 if (!fa->buffer || !fb->buffer) return;
|
10664
|
357
|
32702
|
358 /* Shortcut for duplicate fields (e.g. from RFF flag) */
|
|
359 if (fa->buffer == fb->buffer && pa == pb) {
|
|
360 memset(dest, 0, c->metric_len * sizeof(int));
|
|
361 return;
|
|
362 }
|
10664
|
363
|
32702
|
364 a = fa->buffer->planes[mp] + pa * c->stride[mp] + c->metric_offset;
|
|
365 b = fb->buffer->planes[mp] + pb * c->stride[mp] + c->metric_offset;
|
10664
|
366
|
32702
|
367 for (y = c->metric_h; y; y--) {
|
|
368 for (x = 0; x < w; x += xstep) {
|
|
369 *dest++ = func(a + x, b + x, s);
|
|
370 }
|
|
371 a += ystep; b += ystep;
|
|
372 }
|
10664
|
373 }
|
|
374
|
|
375
|
|
376
|
|
377
|
|
378
|
|
379 static void alloc_metrics(struct pullup_context *c, struct pullup_field *f)
|
|
380 {
|
32702
|
381 f->diffs = calloc(c->metric_len, sizeof(int));
|
|
382 f->comb = calloc(c->metric_len, sizeof(int));
|
|
383 f->var = calloc(c->metric_len, sizeof(int));
|
|
384 /* add more metrics here as needed */
|
10664
|
385 }
|
|
386
|
|
387 static struct pullup_field *make_field_queue(struct pullup_context *c, int len)
|
|
388 {
|
32702
|
389 struct pullup_field *head, *f;
|
|
390 f = head = calloc(1, sizeof(struct pullup_field));
|
|
391 alloc_metrics(c, f);
|
|
392 for (; len > 0; len--) {
|
|
393 f->next = calloc(1, sizeof(struct pullup_field));
|
|
394 f->next->prev = f;
|
|
395 f = f->next;
|
|
396 alloc_metrics(c, f);
|
|
397 }
|
|
398 f->next = head;
|
|
399 head->prev = f;
|
|
400 return head;
|
10664
|
401 }
|
|
402
|
|
403 static void check_field_queue(struct pullup_context *c)
|
|
404 {
|
32702
|
405 if (c->head->next == c->first) {
|
|
406 struct pullup_field *f = calloc(1, sizeof(struct pullup_field));
|
|
407 alloc_metrics(c, f);
|
|
408 f->prev = c->head;
|
|
409 f->next = c->first;
|
|
410 c->head->next = f;
|
|
411 c->first->prev = f;
|
|
412 }
|
10664
|
413 }
|
|
414
|
11412
|
415 void pullup_submit_field(struct pullup_context *c, struct pullup_buffer *b, int parity)
|
10664
|
416 {
|
32702
|
417 struct pullup_field *f;
|
29263
|
418
|
32702
|
419 /* Grow the circular list if needed */
|
|
420 check_field_queue(c);
|
29263
|
421
|
32702
|
422 /* Cannot have two fields of same parity in a row; drop the new one */
|
|
423 if (c->last && c->last->parity == parity) return;
|
10664
|
424
|
32702
|
425 f = c->head;
|
|
426 f->parity = parity;
|
|
427 f->buffer = pullup_lock_buffer(b, parity);
|
|
428 f->flags = 0;
|
|
429 f->breaks = 0;
|
|
430 f->affinity = 0;
|
10664
|
431
|
32702
|
432 compute_metric(c, f, parity, f->prev->prev, parity, c->diff, f->diffs);
|
|
433 compute_metric(c, parity?f->prev:f, 0, parity?f:f->prev, 1, c->comb, f->comb);
|
|
434 compute_metric(c, f, parity, f, -1, c->var, f->var);
|
10664
|
435
|
32702
|
436 /* Advance the circular list */
|
|
437 if (!c->first) c->first = c->head;
|
|
438 c->last = c->head;
|
|
439 c->head = c->head->next;
|
10664
|
440 }
|
|
441
|
|
442 void pullup_flush_fields(struct pullup_context *c)
|
|
443 {
|
32702
|
444 struct pullup_field *f;
|
29263
|
445
|
32702
|
446 for (f = c->first; f && f != c->head; f = f->next) {
|
|
447 pullup_release_buffer(f->buffer, f->parity);
|
|
448 f->buffer = 0;
|
|
449 }
|
|
450 c->first = c->last = 0;
|
10664
|
451 }
|
|
452
|
|
453
|
|
454
|
|
455
|
|
456
|
|
457
|
|
458
|
|
459
|
|
460 #define F_HAVE_BREAKS 1
|
|
461 #define F_HAVE_AFFINITY 2
|
|
462
|
|
463
|
|
464 #define BREAK_LEFT 1
|
|
465 #define BREAK_RIGHT 2
|
|
466
|
|
467
|
|
468
|
|
469
|
|
470 static int queue_length(struct pullup_field *begin, struct pullup_field *end)
|
|
471 {
|
32702
|
472 int count = 1;
|
|
473 struct pullup_field *f;
|
29263
|
474
|
32702
|
475 if (!begin || !end) return 0;
|
|
476 for (f = begin; f != end; f = f->next) count++;
|
|
477 return count;
|
10664
|
478 }
|
|
479
|
|
480 static int find_first_break(struct pullup_field *f, int max)
|
|
481 {
|
32702
|
482 int i;
|
|
483 for (i = 0; i < max; i++) {
|
|
484 if (f->breaks & BREAK_RIGHT || f->next->breaks & BREAK_LEFT)
|
|
485 return i+1;
|
|
486 f = f->next;
|
|
487 }
|
|
488 return 0;
|
10664
|
489 }
|
|
490
|
|
491 static void compute_breaks(struct pullup_context *c, struct pullup_field *f0)
|
|
492 {
|
32702
|
493 int i;
|
|
494 struct pullup_field *f1 = f0->next;
|
|
495 struct pullup_field *f2 = f1->next;
|
|
496 struct pullup_field *f3 = f2->next;
|
|
497 int l, max_l=0, max_r=0;
|
|
498 //struct pullup_field *ff;
|
|
499 //for (i=0, ff=c->first; ff != f0; i++, ff=ff->next);
|
10664
|
500
|
32702
|
501 if (f0->flags & F_HAVE_BREAKS) return;
|
|
502 //printf("\n%d: ", i);
|
|
503 f0->flags |= F_HAVE_BREAKS;
|
10664
|
504
|
32702
|
505 /* Special case when fields are 100% identical */
|
|
506 if (f0->buffer == f2->buffer && f1->buffer != f3->buffer) {
|
|
507 f2->breaks |= BREAK_RIGHT;
|
|
508 return;
|
|
509 }
|
|
510 if (f0->buffer != f2->buffer && f1->buffer == f3->buffer) {
|
|
511 f1->breaks |= BREAK_LEFT;
|
|
512 return;
|
|
513 }
|
10664
|
514
|
32702
|
515 for (i = 0; i < c->metric_len; i++) {
|
|
516 l = f2->diffs[i] - f3->diffs[i];
|
|
517 if (l > max_l) max_l = l;
|
|
518 if (-l > max_r) max_r = -l;
|
|
519 }
|
|
520 /* Don't get tripped up when differences are mostly quant error */
|
|
521 //printf("%d %d\n", max_l, max_r);
|
|
522 if (max_l + max_r < 128) return;
|
|
523 if (max_l > 4*max_r) f1->breaks |= BREAK_LEFT;
|
|
524 if (max_r > 4*max_l) f2->breaks |= BREAK_RIGHT;
|
10664
|
525 }
|
|
526
|
|
527 static void compute_affinity(struct pullup_context *c, struct pullup_field *f)
|
|
528 {
|
32702
|
529 int i;
|
|
530 int max_l=0, max_r=0, l;
|
|
531 if (f->flags & F_HAVE_AFFINITY) return;
|
|
532 f->flags |= F_HAVE_AFFINITY;
|
|
533 if (f->buffer == f->next->next->buffer) {
|
|
534 f->affinity = 1;
|
|
535 f->next->affinity = 0;
|
|
536 f->next->next->affinity = -1;
|
|
537 f->next->flags |= F_HAVE_AFFINITY;
|
|
538 f->next->next->flags |= F_HAVE_AFFINITY;
|
|
539 return;
|
|
540 }
|
|
541 if (1) {
|
|
542 for (i = 0; i < c->metric_len; i++) {
|
|
543 int lv = f->prev->var[i];
|
|
544 int rv = f->next->var[i];
|
|
545 int v = f->var[i];
|
|
546 int lc = f->comb[i] - (v+lv) + ABS(v-lv);
|
|
547 int rc = f->next->comb[i] - (v+rv) + ABS(v-rv);
|
|
548 lc = lc>0 ? lc : 0;
|
|
549 rc = rc>0 ? rc : 0;
|
|
550 l = lc - rc;
|
|
551 if (l > max_l) max_l = l;
|
|
552 if (-l > max_r) max_r = -l;
|
|
553 }
|
|
554 if (max_l + max_r < 64) return;
|
|
555 if (max_r > 6*max_l) f->affinity = -1;
|
|
556 else if (max_l > 6*max_r) f->affinity = 1;
|
|
557 } else {
|
|
558 for (i = 0; i < c->metric_len; i++) {
|
|
559 l = f->comb[i] - f->next->comb[i];
|
|
560 if (l > max_l) max_l = l;
|
|
561 if (-l > max_r) max_r = -l;
|
|
562 }
|
|
563 if (max_l + max_r < 64) return;
|
|
564 if (max_r > 2*max_l) f->affinity = -1;
|
|
565 else if (max_l > 2*max_r) f->affinity = 1;
|
|
566 }
|
10664
|
567 }
|
|
568
|
|
569 static void foo(struct pullup_context *c)
|
|
570 {
|
32702
|
571 struct pullup_field *f = c->first;
|
|
572 int i, n = queue_length(f, c->last);
|
|
573 for (i = 0; i < n-1; i++) {
|
|
574 if (i < n-3) compute_breaks(c, f);
|
|
575 compute_affinity(c, f);
|
|
576 f = f->next;
|
|
577 }
|
10664
|
578 }
|
|
579
|
|
580 static int decide_frame_length(struct pullup_context *c)
|
|
581 {
|
32702
|
582 struct pullup_field *f0 = c->first;
|
|
583 struct pullup_field *f1 = f0->next;
|
|
584 struct pullup_field *f2 = f1->next;
|
|
585 int l;
|
29263
|
586
|
32702
|
587 if (queue_length(c->first, c->last) < 4) return 0;
|
|
588 foo(c);
|
10664
|
589
|
32702
|
590 if (f0->affinity == -1) return 1;
|
10730
|
591
|
32702
|
592 l = find_first_break(f0, 3);
|
|
593 if (l == 1 && c->strict_breaks < 0) l = 0;
|
29263
|
594
|
32702
|
595 switch (l) {
|
|
596 case 1:
|
|
597 if (c->strict_breaks < 1 && f0->affinity == 1 && f1->affinity == -1)
|
|
598 return 2;
|
|
599 else return 1;
|
|
600 case 2:
|
|
601 /* FIXME: strictly speaking, f0->prev is no longer valid... :) */
|
|
602 if (c->strict_pairs
|
|
603 && (f0->prev->breaks & BREAK_RIGHT) && (f2->breaks & BREAK_LEFT)
|
|
604 && (f0->affinity != 1 || f1->affinity != -1) )
|
|
605 return 1;
|
|
606 if (f1->affinity == 1) return 1;
|
|
607 else return 2;
|
|
608 case 3:
|
|
609 if (f2->affinity == 1) return 2;
|
|
610 else return 3;
|
|
611 default:
|
|
612 /* 9 possibilities covered before switch */
|
|
613 if (f1->affinity == 1) return 1; /* covers 6 */
|
|
614 else if (f1->affinity == -1) return 2; /* covers 6 */
|
|
615 else if (f2->affinity == -1) { /* covers 2 */
|
|
616 if (f0->affinity == 1) return 3;
|
|
617 else return 1;
|
|
618 }
|
|
619 else return 2; /* the remaining 6 */
|
|
620 }
|
10664
|
621 }
|
|
622
|
|
623
|
|
624 static void print_aff_and_breaks(struct pullup_context *c, struct pullup_field *f)
|
|
625 {
|
32702
|
626 int i;
|
|
627 struct pullup_field *f0 = f;
|
|
628 const char aff_l[] = "+..", aff_r[] = "..+";
|
|
629 printf("\naffinity: ");
|
|
630 for (i = 0; i < 4; i++) {
|
|
631 printf("%c%d%c", aff_l[1+f->affinity], i, aff_r[1+f->affinity]);
|
|
632 f = f->next;
|
|
633 }
|
|
634 f = f0;
|
|
635 printf("\nbreaks: ");
|
|
636 for (i=0; i<4; i++) {
|
|
637 printf("%c%d%c", f->breaks & BREAK_LEFT ? '|' : '.', i, f->breaks & BREAK_RIGHT ? '|' : '.');
|
|
638 f = f->next;
|
|
639 }
|
|
640 printf("\n");
|
10664
|
641 }
|
|
642
|
|
643
|
|
644
|
|
645
|
|
646
|
|
647 struct pullup_frame *pullup_get_frame(struct pullup_context *c)
|
|
648 {
|
32702
|
649 int i;
|
|
650 struct pullup_frame *fr = c->frame;
|
|
651 int n = decide_frame_length(c);
|
|
652 int aff = c->first->next->affinity;
|
10664
|
653
|
32702
|
654 if (!n) return 0;
|
|
655 if (fr->lock) return 0;
|
10664
|
656
|
32702
|
657 if (c->verbose) {
|
|
658 print_aff_and_breaks(c, c->first);
|
|
659 printf("duration: %d \n", n);
|
|
660 }
|
10664
|
661
|
32702
|
662 fr->lock++;
|
|
663 fr->length = n;
|
|
664 fr->parity = c->first->parity;
|
|
665 fr->buffer = 0;
|
|
666 for (i = 0; i < n; i++) {
|
|
667 /* We cheat and steal the buffer without release+relock */
|
|
668 fr->ifields[i] = c->first->buffer;
|
|
669 c->first->buffer = 0;
|
|
670 c->first = c->first->next;
|
|
671 }
|
29263
|
672
|
32702
|
673 if (n == 1) {
|
|
674 fr->ofields[fr->parity] = fr->ifields[0];
|
|
675 fr->ofields[fr->parity^1] = 0;
|
|
676 } else if (n == 2) {
|
|
677 fr->ofields[fr->parity] = fr->ifields[0];
|
|
678 fr->ofields[fr->parity^1] = fr->ifields[1];
|
|
679 } else if (n == 3) {
|
|
680 if (aff == 0)
|
|
681 aff = (fr->ifields[0] == fr->ifields[1]) ? -1 : 1;
|
|
682 /* else if (c->verbose) printf("forced aff: %d \n", aff); */
|
|
683 fr->ofields[fr->parity] = fr->ifields[1+aff];
|
|
684 fr->ofields[fr->parity^1] = fr->ifields[1];
|
|
685 }
|
|
686 pullup_lock_buffer(fr->ofields[0], 0);
|
|
687 pullup_lock_buffer(fr->ofields[1], 1);
|
29263
|
688
|
32702
|
689 if (fr->ofields[0] == fr->ofields[1]) {
|
|
690 fr->buffer = fr->ofields[0];
|
|
691 pullup_lock_buffer(fr->buffer, 2);
|
|
692 return fr;
|
|
693 }
|
|
694 return fr;
|
10664
|
695 }
|
|
696
|
|
697 static void copy_field(struct pullup_context *c, struct pullup_buffer *dest,
|
32702
|
698 struct pullup_buffer *src, int parity)
|
10664
|
699 {
|
32702
|
700 int i, j;
|
|
701 unsigned char *d, *s;
|
|
702 for (i = 0; i < c->nplanes; i++) {
|
|
703 s = src->planes[i] + parity*c->stride[i];
|
|
704 d = dest->planes[i] + parity*c->stride[i];
|
|
705 for (j = c->h[i]>>1; j; j--) {
|
|
706 memcpy(d, s, c->stride[i]);
|
|
707 s += c->stride[i]<<1;
|
|
708 d += c->stride[i]<<1;
|
|
709 }
|
|
710 }
|
10664
|
711 }
|
|
712
|
|
713 void pullup_pack_frame(struct pullup_context *c, struct pullup_frame *fr)
|
|
714 {
|
32702
|
715 int i;
|
|
716 if (fr->buffer) return;
|
|
717 if (fr->length < 2) return; /* FIXME: deal with this */
|
|
718 for (i = 0; i < 2; i++)
|
|
719 {
|
|
720 if (fr->ofields[i]->lock[i^1]) continue;
|
|
721 fr->buffer = fr->ofields[i];
|
|
722 pullup_lock_buffer(fr->buffer, 2);
|
|
723 copy_field(c, fr->buffer, fr->ofields[i^1], i^1);
|
|
724 return;
|
|
725 }
|
|
726 fr->buffer = pullup_get_buffer(c, 2);
|
|
727 copy_field(c, fr->buffer, fr->ofields[0], 0);
|
|
728 copy_field(c, fr->buffer, fr->ofields[1], 1);
|
10664
|
729 }
|
|
730
|
|
731 void pullup_release_frame(struct pullup_frame *fr)
|
|
732 {
|
32702
|
733 int i;
|
|
734 for (i = 0; i < fr->length; i++)
|
|
735 pullup_release_buffer(fr->ifields[i], fr->parity ^ (i&1));
|
|
736 pullup_release_buffer(fr->ofields[0], 0);
|
|
737 pullup_release_buffer(fr->ofields[1], 1);
|
|
738 if (fr->buffer) pullup_release_buffer(fr->buffer, 2);
|
|
739 fr->lock--;
|
10664
|
740 }
|
|
741
|
|
742
|
|
743
|
|
744
|
|
745
|
|
746
|
17566
|
747 struct pullup_context *pullup_alloc_context(void)
|
10664
|
748 {
|
32702
|
749 struct pullup_context *c;
|
10664
|
750
|
32702
|
751 c = calloc(1, sizeof(struct pullup_context));
|
10664
|
752
|
32702
|
753 return c;
|
10664
|
754 }
|
|
755
|
|
756 void pullup_preinit_context(struct pullup_context *c)
|
|
757 {
|
32702
|
758 c->bpp = calloc(c->nplanes, sizeof(int));
|
|
759 c->w = calloc(c->nplanes, sizeof(int));
|
|
760 c->h = calloc(c->nplanes, sizeof(int));
|
|
761 c->stride = calloc(c->nplanes, sizeof(int));
|
|
762 c->background = calloc(c->nplanes, sizeof(int));
|
10664
|
763 }
|
|
764
|
|
765 void pullup_init_context(struct pullup_context *c)
|
|
766 {
|
32702
|
767 int mp = c->metric_plane;
|
|
768 if (c->nbuffers < 10) c->nbuffers = 10;
|
|
769 c->buffers = calloc(c->nbuffers, sizeof (struct pullup_buffer));
|
10664
|
770
|
32702
|
771 c->metric_w = (c->w[mp] - ((c->junk_left + c->junk_right) << 3)) >> 3;
|
|
772 c->metric_h = (c->h[mp] - ((c->junk_top + c->junk_bottom) << 1)) >> 3;
|
|
773 c->metric_offset = c->junk_left*c->bpp[mp] + (c->junk_top<<1)*c->stride[mp];
|
|
774 c->metric_len = c->metric_w * c->metric_h;
|
29263
|
775
|
32702
|
776 c->head = make_field_queue(c, 8);
|
10664
|
777
|
32702
|
778 c->frame = calloc(1, sizeof (struct pullup_frame));
|
|
779 c->frame->ifields = calloc(3, sizeof (struct pullup_buffer *));
|
10664
|
780
|
32702
|
781 switch(c->format) {
|
|
782 case PULLUP_FMT_Y:
|
|
783 c->diff = diff_y;
|
|
784 c->comb = licomb_y;
|
|
785 c->var = var_y;
|
28290
|
786 #if ARCH_X86
|
|
787 #if HAVE_MMX
|
32702
|
788 if (c->cpu & PULLUP_CPU_MMX) {
|
|
789 c->diff = diff_y_mmx;
|
|
790 c->comb = licomb_y_mmx;
|
|
791 c->var = var_y_mmx;
|
|
792 }
|
10664
|
793 #endif
|
13720
|
794 #endif
|
32702
|
795 /* c->comb = qpcomb_y; */
|
|
796 break;
|
10664
|
797 #if 0
|
32702
|
798 case PULLUP_FMT_YUY2:
|
|
799 c->diff = diff_yuy2;
|
|
800 break;
|
|
801 case PULLUP_FMT_RGB32:
|
|
802 c->diff = diff_rgb32;
|
|
803 break;
|
10664
|
804 #endif
|
32702
|
805 }
|
10664
|
806 }
|
|
807
|
|
808 void pullup_free_context(struct pullup_context *c)
|
|
809 {
|
32702
|
810 struct pullup_field *f;
|
|
811 free(c->buffers);
|
|
812 f = c->head;
|
|
813 do {
|
|
814 if (!f) break;
|
|
815 free(f->diffs);
|
|
816 free(f->comb);
|
|
817 f = f->next;
|
|
818 free(f->prev);
|
|
819 } while (f != c->head);
|
|
820 free(c->frame);
|
|
821 free(c);
|
10664
|
822 }
|