2 * This file is part of Libav.
4 * Libav is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * Libav 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 GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with Libav; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "libavutil/attributes.h"
21 #include "copy_block.h"
22 #include "simple_idct.h"
24 #include "mpegvideo.h"
27 uint32_t ff_square_tab[512] = { 0, };
29 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
33 uint32_t *sq = ff_square_tab + 256;
35 for (i = 0; i < h; i++) {
36 s += sq[pix1[0] - pix2[0]];
37 s += sq[pix1[1] - pix2[1]];
38 s += sq[pix1[2] - pix2[2]];
39 s += sq[pix1[3] - pix2[3]];
46 static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
50 uint32_t *sq = ff_square_tab + 256;
52 for (i = 0; i < h; i++) {
53 s += sq[pix1[0] - pix2[0]];
54 s += sq[pix1[1] - pix2[1]];
55 s += sq[pix1[2] - pix2[2]];
56 s += sq[pix1[3] - pix2[3]];
57 s += sq[pix1[4] - pix2[4]];
58 s += sq[pix1[5] - pix2[5]];
59 s += sq[pix1[6] - pix2[6]];
60 s += sq[pix1[7] - pix2[7]];
67 static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
71 uint32_t *sq = ff_square_tab + 256;
73 for (i = 0; i < h; i++) {
74 s += sq[pix1[0] - pix2[0]];
75 s += sq[pix1[1] - pix2[1]];
76 s += sq[pix1[2] - pix2[2]];
77 s += sq[pix1[3] - pix2[3]];
78 s += sq[pix1[4] - pix2[4]];
79 s += sq[pix1[5] - pix2[5]];
80 s += sq[pix1[6] - pix2[6]];
81 s += sq[pix1[7] - pix2[7]];
82 s += sq[pix1[8] - pix2[8]];
83 s += sq[pix1[9] - pix2[9]];
84 s += sq[pix1[10] - pix2[10]];
85 s += sq[pix1[11] - pix2[11]];
86 s += sq[pix1[12] - pix2[12]];
87 s += sq[pix1[13] - pix2[13]];
88 s += sq[pix1[14] - pix2[14]];
89 s += sq[pix1[15] - pix2[15]];
97 static int sum_abs_dctelem_c(int16_t *block)
101 for (i = 0; i < 64; i++)
102 sum += FFABS(block[i]);
106 #define avg2(a, b) ((a + b + 1) >> 1)
107 #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)
109 static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
110 int line_size, int h)
114 for (i = 0; i < h; i++) {
115 s += abs(pix1[0] - pix2[0]);
116 s += abs(pix1[1] - pix2[1]);
117 s += abs(pix1[2] - pix2[2]);
118 s += abs(pix1[3] - pix2[3]);
119 s += abs(pix1[4] - pix2[4]);
120 s += abs(pix1[5] - pix2[5]);
121 s += abs(pix1[6] - pix2[6]);
122 s += abs(pix1[7] - pix2[7]);
123 s += abs(pix1[8] - pix2[8]);
124 s += abs(pix1[9] - pix2[9]);
125 s += abs(pix1[10] - pix2[10]);
126 s += abs(pix1[11] - pix2[11]);
127 s += abs(pix1[12] - pix2[12]);
128 s += abs(pix1[13] - pix2[13]);
129 s += abs(pix1[14] - pix2[14]);
130 s += abs(pix1[15] - pix2[15]);
137 static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
138 int line_size, int h)
142 for (i = 0; i < h; i++) {
143 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
144 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
145 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
146 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
147 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
148 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
149 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
150 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
151 s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
152 s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
153 s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
154 s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
155 s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
156 s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
157 s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
158 s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
165 static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
166 int line_size, int h)
169 uint8_t *pix3 = pix2 + line_size;
171 for (i = 0; i < h; i++) {
172 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
173 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
174 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
175 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
176 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
177 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
178 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
179 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
180 s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
181 s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
182 s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
183 s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
184 s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
185 s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
186 s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
187 s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
195 static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
196 int line_size, int h)
199 uint8_t *pix3 = pix2 + line_size;
201 for (i = 0; i < h; i++) {
202 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
203 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
204 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
205 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
206 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
207 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
208 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
209 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
210 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
211 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
212 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
213 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
214 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
215 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
216 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
217 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
225 static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
226 int line_size, int h)
230 for (i = 0; i < h; i++) {
231 s += abs(pix1[0] - pix2[0]);
232 s += abs(pix1[1] - pix2[1]);
233 s += abs(pix1[2] - pix2[2]);
234 s += abs(pix1[3] - pix2[3]);
235 s += abs(pix1[4] - pix2[4]);
236 s += abs(pix1[5] - pix2[5]);
237 s += abs(pix1[6] - pix2[6]);
238 s += abs(pix1[7] - pix2[7]);
245 static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
246 int line_size, int h)
250 for (i = 0; i < h; i++) {
251 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
252 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
253 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
254 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
255 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
256 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
257 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
258 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
265 static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
266 int line_size, int h)
269 uint8_t *pix3 = pix2 + line_size;
271 for (i = 0; i < h; i++) {
272 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
273 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
274 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
275 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
276 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
277 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
278 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
279 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
287 static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
288 int line_size, int h)
291 uint8_t *pix3 = pix2 + line_size;
293 for (i = 0; i < h; i++) {
294 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
295 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
296 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
297 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
298 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
299 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
300 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
301 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
309 static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
311 int score1 = 0, score2 = 0, x, y;
313 for (y = 0; y < h; y++) {
314 for (x = 0; x < 16; x++)
315 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
317 for (x = 0; x < 15; x++)
318 score2 += FFABS(s1[x] - s1[x + stride] -
319 s1[x + 1] + s1[x + stride + 1]) -
320 FFABS(s2[x] - s2[x + stride] -
321 s2[x + 1] + s2[x + stride + 1]);
328 return score1 + FFABS(score2) * c->avctx->nsse_weight;
330 return score1 + FFABS(score2) * 8;
333 static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
335 int score1 = 0, score2 = 0, x, y;
337 for (y = 0; y < h; y++) {
338 for (x = 0; x < 8; x++)
339 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
341 for (x = 0; x < 7; x++)
342 score2 += FFABS(s1[x] - s1[x + stride] -
343 s1[x + 1] + s1[x + stride + 1]) -
344 FFABS(s2[x] - s2[x + stride] -
345 s2[x + 1] + s2[x + stride + 1]);
352 return score1 + FFABS(score2) * c->avctx->nsse_weight;
354 return score1 + FFABS(score2) * 8;
357 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
363 void ff_set_cmp(MECmpContext *c, me_cmp_func *cmp, int type)
367 memset(cmp, 0, sizeof(void *) * 6);
369 for (i = 0; i < 6; i++) {
370 switch (type & 0xFF) {
375 cmp[i] = c->hadamard8_diff[i];
381 cmp[i] = c->dct_sad[i];
384 cmp[i] = c->dct264_sad[i];
387 cmp[i] = c->dct_max[i];
390 cmp[i] = c->quant_psnr[i];
411 av_log(NULL, AV_LOG_ERROR,
412 "internal error in cmp function selection\n");
417 #define BUTTERFLY2(o1, o2, i1, i2) \
421 #define BUTTERFLY1(x, y) \
430 #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))
432 static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,
433 uint8_t *src, int stride, int h)
435 int i, temp[64], sum = 0;
439 for (i = 0; i < 8; i++) {
440 // FIXME: try pointer walks
441 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
442 src[stride * i + 0] - dst[stride * i + 0],
443 src[stride * i + 1] - dst[stride * i + 1]);
444 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
445 src[stride * i + 2] - dst[stride * i + 2],
446 src[stride * i + 3] - dst[stride * i + 3]);
447 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
448 src[stride * i + 4] - dst[stride * i + 4],
449 src[stride * i + 5] - dst[stride * i + 5]);
450 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
451 src[stride * i + 6] - dst[stride * i + 6],
452 src[stride * i + 7] - dst[stride * i + 7]);
454 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
455 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
456 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
457 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
459 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
460 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
461 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
462 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
465 for (i = 0; i < 8; i++) {
466 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
467 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
468 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
469 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
471 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
472 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
473 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
474 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
476 sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +
477 BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +
478 BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +
479 BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
484 static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,
485 uint8_t *dummy, int stride, int h)
487 int i, temp[64], sum = 0;
491 for (i = 0; i < 8; i++) {
492 // FIXME: try pointer walks
493 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
494 src[stride * i + 0], src[stride * i + 1]);
495 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
496 src[stride * i + 2], src[stride * i + 3]);
497 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
498 src[stride * i + 4], src[stride * i + 5]);
499 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
500 src[stride * i + 6], src[stride * i + 7]);
502 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
503 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
504 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
505 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
507 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
508 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
509 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
510 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
513 for (i = 0; i < 8; i++) {
514 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
515 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
516 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
517 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
519 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
520 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
521 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
522 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
525 BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])
526 + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])
527 + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])
528 + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
531 sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean
536 static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
537 uint8_t *src2, int stride, int h)
539 LOCAL_ALIGNED_16(int16_t, temp, [64]);
543 s->pdsp.diff_pixels(temp, src1, src2, stride);
545 return s->mecc.sum_abs_dctelem(temp);
551 const int s07 = SRC(0) + SRC(7); \
552 const int s16 = SRC(1) + SRC(6); \
553 const int s25 = SRC(2) + SRC(5); \
554 const int s34 = SRC(3) + SRC(4); \
555 const int a0 = s07 + s34; \
556 const int a1 = s16 + s25; \
557 const int a2 = s07 - s34; \
558 const int a3 = s16 - s25; \
559 const int d07 = SRC(0) - SRC(7); \
560 const int d16 = SRC(1) - SRC(6); \
561 const int d25 = SRC(2) - SRC(5); \
562 const int d34 = SRC(3) - SRC(4); \
563 const int a4 = d16 + d25 + (d07 + (d07 >> 1)); \
564 const int a5 = d07 - d34 - (d25 + (d25 >> 1)); \
565 const int a6 = d07 + d34 - (d16 + (d16 >> 1)); \
566 const int a7 = d16 - d25 + (d34 + (d34 >> 1)); \
568 DST(1, a4 + (a7 >> 2)); \
569 DST(2, a2 + (a3 >> 1)); \
570 DST(3, a5 + (a6 >> 2)); \
572 DST(5, a6 - (a5 >> 2)); \
573 DST(6, (a2 >> 1) - a3); \
574 DST(7, (a4 >> 2) - a7); \
577 static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
578 uint8_t *src2, int stride, int h)
583 s->pdsp.diff_pixels(dct[0], src1, src2, stride);
585 #define SRC(x) dct[i][x]
586 #define DST(x, v) dct[i][x] = v
587 for (i = 0; i < 8; i++)
592 #define SRC(x) dct[x][i]
593 #define DST(x, v) sum += FFABS(v)
594 for (i = 0; i < 8; i++)
602 static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
603 uint8_t *src2, int stride, int h)
605 LOCAL_ALIGNED_16(int16_t, temp, [64]);
610 s->pdsp.diff_pixels(temp, src1, src2, stride);
613 for (i = 0; i < 64; i++)
614 sum = FFMAX(sum, FFABS(temp[i]));
619 static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
620 uint8_t *src2, int stride, int h)
622 LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);
623 int16_t *const bak = temp + 64;
629 s->pdsp.diff_pixels(temp, src1, src2, stride);
631 memcpy(bak, temp, 64 * sizeof(int16_t));
633 s->block_last_index[0 /* FIXME */] =
634 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
635 s->dct_unquantize_inter(s, temp, 0, s->qscale);
636 ff_simple_idct_8(temp); // FIXME
638 for (i = 0; i < 64; i++)
639 sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);
644 static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
647 const uint8_t *scantable = s->intra_scantable.permutated;
648 LOCAL_ALIGNED_16(int16_t, temp, [64]);
649 LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);
650 LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);
651 int i, last, run, bits, level, distortion, start_i;
652 const int esc_length = s->ac_esc_length;
653 uint8_t *length, *last_length;
657 copy_block8(lsrc1, src1, 8, stride, 8);
658 copy_block8(lsrc2, src2, 8, stride, 8);
660 s->pdsp.diff_pixels(temp, lsrc1, lsrc2, 8);
662 s->block_last_index[0 /* FIXME */] =
664 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
670 length = s->intra_ac_vlc_length;
671 last_length = s->intra_ac_vlc_last_length;
672 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
675 length = s->inter_ac_vlc_length;
676 last_length = s->inter_ac_vlc_last_length;
679 if (last >= start_i) {
681 for (i = start_i; i < last; i++) {
682 int j = scantable[i];
687 if ((level & (~127)) == 0)
688 bits += length[UNI_AC_ENC_INDEX(run, level)];
697 level = temp[i] + 64;
701 if ((level & (~127)) == 0) {
702 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
709 s->dct_unquantize_intra(s, temp, 0, s->qscale);
711 s->dct_unquantize_inter(s, temp, 0, s->qscale);
714 s->idsp.idct_add(lsrc2, 8, temp);
716 distortion = s->mecc.sse[1](NULL, lsrc2, lsrc1, 8, 8);
718 return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);
721 static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
724 const uint8_t *scantable = s->intra_scantable.permutated;
725 LOCAL_ALIGNED_16(int16_t, temp, [64]);
726 int i, last, run, bits, level, start_i;
727 const int esc_length = s->ac_esc_length;
728 uint8_t *length, *last_length;
732 s->pdsp.diff_pixels(temp, src1, src2, stride);
734 s->block_last_index[0 /* FIXME */] =
736 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
742 length = s->intra_ac_vlc_length;
743 last_length = s->intra_ac_vlc_last_length;
744 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
747 length = s->inter_ac_vlc_length;
748 last_length = s->inter_ac_vlc_last_length;
751 if (last >= start_i) {
753 for (i = start_i; i < last; i++) {
754 int j = scantable[i];
759 if ((level & (~127)) == 0)
760 bits += length[UNI_AC_ENC_INDEX(run, level)];
769 level = temp[i] + 64;
773 if ((level & (~127)) == 0)
774 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
782 #define VSAD_INTRA(size) \
783 static int vsad_intra ## size ## _c(MpegEncContext *c, \
784 uint8_t *s, uint8_t *dummy, \
787 int score = 0, x, y; \
789 for (y = 1; y < h; y++) { \
790 for (x = 0; x < size; x += 4) { \
791 score += FFABS(s[x] - s[x + stride]) + \
792 FFABS(s[x + 1] - s[x + stride + 1]) + \
793 FFABS(s[x + 2] - s[x + 2 + stride]) + \
794 FFABS(s[x + 3] - s[x + 3 + stride]); \
804 static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
809 for (y = 1; y < h; y++) {
810 for (x = 0; x < 16; x++)
811 score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);
819 #define SQ(a) ((a) * (a))
820 #define VSSE_INTRA(size) \
821 static int vsse_intra ## size ## _c(MpegEncContext *c, \
822 uint8_t *s, uint8_t *dummy, \
825 int score = 0, x, y; \
827 for (y = 1; y < h; y++) { \
828 for (x = 0; x < size; x += 4) { \
829 score += SQ(s[x] - s[x + stride]) + \
830 SQ(s[x + 1] - s[x + stride + 1]) + \
831 SQ(s[x + 2] - s[x + stride + 2]) + \
832 SQ(s[x + 3] - s[x + stride + 3]); \
842 static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
847 for (y = 1; y < h; y++) {
848 for (x = 0; x < 16; x++)
849 score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);
857 #define WRAPPER8_16_SQ(name8, name16) \
858 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src, \
863 score += name8(s, dst, src, stride, 8); \
864 score += name8(s, dst + 8, src + 8, stride, 8); \
868 score += name8(s, dst, src, stride, 8); \
869 score += name8(s, dst + 8, src + 8, stride, 8); \
874 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)
875 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)
876 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)
878 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)
880 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)
881 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
882 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
883 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
885 av_cold void ff_me_cmp_init_static(void)
889 for (i = 0; i < 512; i++)
890 ff_square_tab[i] = (i - 256) * (i - 256);
893 av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)
895 c->sum_abs_dctelem = sum_abs_dctelem_c;
897 /* TODO [0] 16 [1] 8 */
898 c->pix_abs[0][0] = pix_abs16_c;
899 c->pix_abs[0][1] = pix_abs16_x2_c;
900 c->pix_abs[0][2] = pix_abs16_y2_c;
901 c->pix_abs[0][3] = pix_abs16_xy2_c;
902 c->pix_abs[1][0] = pix_abs8_c;
903 c->pix_abs[1][1] = pix_abs8_x2_c;
904 c->pix_abs[1][2] = pix_abs8_y2_c;
905 c->pix_abs[1][3] = pix_abs8_xy2_c;
907 #define SET_CMP_FUNC(name) \
908 c->name[0] = name ## 16_c; \
909 c->name[1] = name ## 8x8_c;
911 SET_CMP_FUNC(hadamard8_diff)
912 c->hadamard8_diff[4] = hadamard8_intra16_c;
913 c->hadamard8_diff[5] = hadamard8_intra8x8_c;
914 SET_CMP_FUNC(dct_sad)
915 SET_CMP_FUNC(dct_max)
917 SET_CMP_FUNC(dct264_sad)
919 c->sad[0] = pix_abs16_c;
920 c->sad[1] = pix_abs8_c;
924 SET_CMP_FUNC(quant_psnr)
927 c->vsad[0] = vsad16_c;
928 c->vsad[4] = vsad_intra16_c;
929 c->vsad[5] = vsad_intra8_c;
930 c->vsse[0] = vsse16_c;
931 c->vsse[4] = vsse_intra16_c;
932 c->vsse[5] = vsse_intra8_c;
933 c->nsse[0] = nsse16_c;
934 c->nsse[1] = nsse8_c;
937 ff_me_cmp_init_arm(c, avctx);
939 ff_me_cmp_init_ppc(c, avctx);
941 ff_me_cmp_init_x86(c, avctx);