3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 #include "libavutil/attributes.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/internal.h"
32 #include "copy_block.h"
35 #include "simple_idct.h"
37 #include "imgconvert.h"
39 #include "mpegvideo.h"
42 uint32_t ff_square_tab[512] = { 0, };
45 #include "dsputilenc_template.c"
49 #include "dsputilenc_template.c"
51 static int pix_sum_c(uint8_t *pix, int line_size)
55 for (i = 0; i < 16; i++) {
56 for (j = 0; j < 16; j += 8) {
67 pix += line_size - 16;
72 static int pix_norm1_c(uint8_t *pix, int line_size)
75 uint32_t *sq = ff_square_tab + 256;
77 for (i = 0; i < 16; i++) {
78 for (j = 0; j < 16; j += 8) {
90 register uint64_t x = *(uint64_t *) pix;
92 s += sq[(x >> 8) & 0xff];
93 s += sq[(x >> 16) & 0xff];
94 s += sq[(x >> 24) & 0xff];
95 s += sq[(x >> 32) & 0xff];
96 s += sq[(x >> 40) & 0xff];
97 s += sq[(x >> 48) & 0xff];
98 s += sq[(x >> 56) & 0xff];
100 register uint32_t x = *(uint32_t *) pix;
102 s += sq[(x >> 8) & 0xff];
103 s += sq[(x >> 16) & 0xff];
104 s += sq[(x >> 24) & 0xff];
105 x = *(uint32_t *) (pix + 4);
107 s += sq[(x >> 8) & 0xff];
108 s += sq[(x >> 16) & 0xff];
109 s += sq[(x >> 24) & 0xff];
114 pix += line_size - 16;
119 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
120 int line_size, int h)
123 uint32_t *sq = ff_square_tab + 256;
125 for (i = 0; i < h; i++) {
126 s += sq[pix1[0] - pix2[0]];
127 s += sq[pix1[1] - pix2[1]];
128 s += sq[pix1[2] - pix2[2]];
129 s += sq[pix1[3] - pix2[3]];
136 static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
137 int line_size, int h)
140 uint32_t *sq = ff_square_tab + 256;
142 for (i = 0; i < h; i++) {
143 s += sq[pix1[0] - pix2[0]];
144 s += sq[pix1[1] - pix2[1]];
145 s += sq[pix1[2] - pix2[2]];
146 s += sq[pix1[3] - pix2[3]];
147 s += sq[pix1[4] - pix2[4]];
148 s += sq[pix1[5] - pix2[5]];
149 s += sq[pix1[6] - pix2[6]];
150 s += sq[pix1[7] - pix2[7]];
157 static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
158 int line_size, int h)
161 uint32_t *sq = ff_square_tab + 256;
163 for (i = 0; i < h; i++) {
164 s += sq[pix1[0] - pix2[0]];
165 s += sq[pix1[1] - pix2[1]];
166 s += sq[pix1[2] - pix2[2]];
167 s += sq[pix1[3] - pix2[3]];
168 s += sq[pix1[4] - pix2[4]];
169 s += sq[pix1[5] - pix2[5]];
170 s += sq[pix1[6] - pix2[6]];
171 s += sq[pix1[7] - pix2[7]];
172 s += sq[pix1[8] - pix2[8]];
173 s += sq[pix1[9] - pix2[9]];
174 s += sq[pix1[10] - pix2[10]];
175 s += sq[pix1[11] - pix2[11]];
176 s += sq[pix1[12] - pix2[12]];
177 s += sq[pix1[13] - pix2[13]];
178 s += sq[pix1[14] - pix2[14]];
179 s += sq[pix1[15] - pix2[15]];
187 static void diff_pixels_c(int16_t *av_restrict block, const uint8_t *s1,
188 const uint8_t *s2, int stride)
192 /* read the pixels */
193 for (i = 0; i < 8; i++) {
194 block[0] = s1[0] - s2[0];
195 block[1] = s1[1] - s2[1];
196 block[2] = s1[2] - s2[2];
197 block[3] = s1[3] - s2[3];
198 block[4] = s1[4] - s2[4];
199 block[5] = s1[5] - s2[5];
200 block[6] = s1[6] - s2[6];
201 block[7] = s1[7] - s2[7];
208 static int sum_abs_dctelem_c(int16_t *block)
212 for (i = 0; i < 64; i++)
213 sum += FFABS(block[i]);
217 #define avg2(a, b) ((a + b + 1) >> 1)
218 #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)
220 static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
221 int line_size, int h)
225 for (i = 0; i < h; i++) {
226 s += abs(pix1[0] - pix2[0]);
227 s += abs(pix1[1] - pix2[1]);
228 s += abs(pix1[2] - pix2[2]);
229 s += abs(pix1[3] - pix2[3]);
230 s += abs(pix1[4] - pix2[4]);
231 s += abs(pix1[5] - pix2[5]);
232 s += abs(pix1[6] - pix2[6]);
233 s += abs(pix1[7] - pix2[7]);
234 s += abs(pix1[8] - pix2[8]);
235 s += abs(pix1[9] - pix2[9]);
236 s += abs(pix1[10] - pix2[10]);
237 s += abs(pix1[11] - pix2[11]);
238 s += abs(pix1[12] - pix2[12]);
239 s += abs(pix1[13] - pix2[13]);
240 s += abs(pix1[14] - pix2[14]);
241 s += abs(pix1[15] - pix2[15]);
248 static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
249 int line_size, int h)
253 for (i = 0; i < h; i++) {
254 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
255 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
256 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
257 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
258 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
259 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
260 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
261 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
262 s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
263 s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
264 s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
265 s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
266 s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
267 s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
268 s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
269 s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
276 static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
277 int line_size, int h)
280 uint8_t *pix3 = pix2 + line_size;
282 for (i = 0; i < h; i++) {
283 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
284 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
285 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
286 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
287 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
288 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
289 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
290 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
291 s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
292 s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
293 s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
294 s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
295 s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
296 s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
297 s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
298 s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
306 static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
307 int line_size, int h)
310 uint8_t *pix3 = pix2 + line_size;
312 for (i = 0; i < h; i++) {
313 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
314 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
315 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
316 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
317 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
318 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
319 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
320 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
321 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
322 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
323 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
324 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
325 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
326 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
327 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
328 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
336 static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
337 int line_size, int h)
341 for (i = 0; i < h; i++) {
342 s += abs(pix1[0] - pix2[0]);
343 s += abs(pix1[1] - pix2[1]);
344 s += abs(pix1[2] - pix2[2]);
345 s += abs(pix1[3] - pix2[3]);
346 s += abs(pix1[4] - pix2[4]);
347 s += abs(pix1[5] - pix2[5]);
348 s += abs(pix1[6] - pix2[6]);
349 s += abs(pix1[7] - pix2[7]);
356 static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
357 int line_size, int h)
361 for (i = 0; i < h; i++) {
362 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
363 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
364 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
365 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
366 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
367 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
368 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
369 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
376 static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
377 int line_size, int h)
380 uint8_t *pix3 = pix2 + line_size;
382 for (i = 0; i < h; i++) {
383 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
384 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
385 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
386 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
387 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
388 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
389 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
390 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
398 static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
399 int line_size, int h)
402 uint8_t *pix3 = pix2 + line_size;
404 for (i = 0; i < h; i++) {
405 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
406 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
407 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
408 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
409 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
410 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
411 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
412 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
420 static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
422 int score1 = 0, score2 = 0, x, y;
424 for (y = 0; y < h; y++) {
425 for (x = 0; x < 16; x++)
426 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
428 for (x = 0; x < 15; x++)
429 score2 += FFABS(s1[x] - s1[x + stride] -
430 s1[x + 1] + s1[x + stride + 1]) -
431 FFABS(s2[x] - s2[x + stride] -
432 s2[x + 1] + s2[x + stride + 1]);
439 return score1 + FFABS(score2) * c->avctx->nsse_weight;
441 return score1 + FFABS(score2) * 8;
444 static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
446 int score1 = 0, score2 = 0, x, y;
448 for (y = 0; y < h; y++) {
449 for (x = 0; x < 8; x++)
450 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
452 for (x = 0; x < 7; x++)
453 score2 += FFABS(s1[x] - s1[x + stride] -
454 s1[x + 1] + s1[x + stride + 1]) -
455 FFABS(s2[x] - s2[x + stride] -
456 s2[x + 1] + s2[x + stride + 1]);
463 return score1 + FFABS(score2) * c->avctx->nsse_weight;
465 return score1 + FFABS(score2) * 8;
468 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],
469 int16_t basis[64], int scale)
472 unsigned int sum = 0;
474 for (i = 0; i < 8 * 8; i++) {
475 int b = rem[i] + ((basis[i] * scale +
476 (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>
477 (BASIS_SHIFT - RECON_SHIFT));
480 av_assert2(-512 < b && b < 512);
482 sum += (w * b) * (w * b) >> 4;
487 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)
491 for (i = 0; i < 8 * 8; i++)
492 rem[i] += (basis[i] * scale +
493 (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>
494 (BASIS_SHIFT - RECON_SHIFT);
497 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
503 void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)
507 memset(cmp, 0, sizeof(void *) * 6);
509 for (i = 0; i < 6; i++) {
510 switch (type & 0xFF) {
515 cmp[i] = c->hadamard8_diff[i];
521 cmp[i] = c->dct_sad[i];
524 cmp[i] = c->dct264_sad[i];
527 cmp[i] = c->dct_max[i];
530 cmp[i] = c->quant_psnr[i];
559 av_log(NULL, AV_LOG_ERROR,
560 "internal error in cmp function selection\n");
565 #define BUTTERFLY2(o1, o2, i1, i2) \
569 #define BUTTERFLY1(x, y) \
578 #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))
580 static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,
581 uint8_t *src, int stride, int h)
583 int i, temp[64], sum = 0;
587 for (i = 0; i < 8; i++) {
588 // FIXME: try pointer walks
589 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
590 src[stride * i + 0] - dst[stride * i + 0],
591 src[stride * i + 1] - dst[stride * i + 1]);
592 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
593 src[stride * i + 2] - dst[stride * i + 2],
594 src[stride * i + 3] - dst[stride * i + 3]);
595 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
596 src[stride * i + 4] - dst[stride * i + 4],
597 src[stride * i + 5] - dst[stride * i + 5]);
598 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
599 src[stride * i + 6] - dst[stride * i + 6],
600 src[stride * i + 7] - dst[stride * i + 7]);
602 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
603 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
604 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
605 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
607 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
608 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
609 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
610 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
613 for (i = 0; i < 8; i++) {
614 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
615 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
616 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
617 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
619 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
620 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
621 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
622 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
624 sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +
625 BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +
626 BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +
627 BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
632 static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,
633 uint8_t *dummy, int stride, int h)
635 int i, temp[64], sum = 0;
639 for (i = 0; i < 8; i++) {
640 // FIXME: try pointer walks
641 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
642 src[stride * i + 0], src[stride * i + 1]);
643 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
644 src[stride * i + 2], src[stride * i + 3]);
645 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
646 src[stride * i + 4], src[stride * i + 5]);
647 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
648 src[stride * i + 6], src[stride * i + 7]);
650 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
651 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
652 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
653 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
655 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
656 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
657 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
658 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
661 for (i = 0; i < 8; i++) {
662 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
663 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
664 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
665 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
667 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
668 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
669 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
670 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
673 BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])
674 + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])
675 + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])
676 + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
679 sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean
684 static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
685 uint8_t *src2, int stride, int h)
687 LOCAL_ALIGNED_16(int16_t, temp, [64]);
691 s->dsp.diff_pixels(temp, src1, src2, stride);
693 return s->dsp.sum_abs_dctelem(temp);
699 const int s07 = SRC(0) + SRC(7); \
700 const int s16 = SRC(1) + SRC(6); \
701 const int s25 = SRC(2) + SRC(5); \
702 const int s34 = SRC(3) + SRC(4); \
703 const int a0 = s07 + s34; \
704 const int a1 = s16 + s25; \
705 const int a2 = s07 - s34; \
706 const int a3 = s16 - s25; \
707 const int d07 = SRC(0) - SRC(7); \
708 const int d16 = SRC(1) - SRC(6); \
709 const int d25 = SRC(2) - SRC(5); \
710 const int d34 = SRC(3) - SRC(4); \
711 const int a4 = d16 + d25 + (d07 + (d07 >> 1)); \
712 const int a5 = d07 - d34 - (d25 + (d25 >> 1)); \
713 const int a6 = d07 + d34 - (d16 + (d16 >> 1)); \
714 const int a7 = d16 - d25 + (d34 + (d34 >> 1)); \
716 DST(1, a4 + (a7 >> 2)); \
717 DST(2, a2 + (a3 >> 1)); \
718 DST(3, a5 + (a6 >> 2)); \
720 DST(5, a6 - (a5 >> 2)); \
721 DST(6, (a2 >> 1) - a3); \
722 DST(7, (a4 >> 2) - a7); \
725 static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
726 uint8_t *src2, int stride, int h)
731 s->dsp.diff_pixels(dct[0], src1, src2, stride);
733 #define SRC(x) dct[i][x]
734 #define DST(x, v) dct[i][x] = v
735 for (i = 0; i < 8; i++)
740 #define SRC(x) dct[x][i]
741 #define DST(x, v) sum += FFABS(v)
742 for (i = 0; i < 8; i++)
750 static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
751 uint8_t *src2, int stride, int h)
753 LOCAL_ALIGNED_16(int16_t, temp, [64]);
758 s->dsp.diff_pixels(temp, src1, src2, stride);
761 for (i = 0; i < 64; i++)
762 sum = FFMAX(sum, FFABS(temp[i]));
767 static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
768 uint8_t *src2, int stride, int h)
770 LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);
771 int16_t *const bak = temp + 64;
777 s->dsp.diff_pixels(temp, src1, src2, stride);
779 memcpy(bak, temp, 64 * sizeof(int16_t));
781 s->block_last_index[0 /* FIXME */] =
782 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
783 s->dct_unquantize_inter(s, temp, 0, s->qscale);
784 ff_simple_idct_8(temp); // FIXME
786 for (i = 0; i < 64; i++)
787 sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);
792 static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
795 const uint8_t *scantable = s->intra_scantable.permutated;
796 LOCAL_ALIGNED_16(int16_t, temp, [64]);
797 LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);
798 LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);
799 int i, last, run, bits, level, distortion, start_i;
800 const int esc_length = s->ac_esc_length;
801 uint8_t *length, *last_length;
805 copy_block8(lsrc1, src1, 8, stride, 8);
806 copy_block8(lsrc2, src2, 8, stride, 8);
808 s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);
810 s->block_last_index[0 /* FIXME */] =
812 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
818 length = s->intra_ac_vlc_length;
819 last_length = s->intra_ac_vlc_last_length;
820 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
823 length = s->inter_ac_vlc_length;
824 last_length = s->inter_ac_vlc_last_length;
827 if (last >= start_i) {
829 for (i = start_i; i < last; i++) {
830 int j = scantable[i];
835 if ((level & (~127)) == 0)
836 bits += length[UNI_AC_ENC_INDEX(run, level)];
845 level = temp[i] + 64;
847 av_assert2(level - 64);
849 if ((level & (~127)) == 0) {
850 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
857 s->dct_unquantize_intra(s, temp, 0, s->qscale);
859 s->dct_unquantize_inter(s, temp, 0, s->qscale);
862 s->idsp.idct_add(lsrc2, 8, temp);
864 distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);
866 return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);
869 static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
872 const uint8_t *scantable = s->intra_scantable.permutated;
873 LOCAL_ALIGNED_16(int16_t, temp, [64]);
874 int i, last, run, bits, level, start_i;
875 const int esc_length = s->ac_esc_length;
876 uint8_t *length, *last_length;
880 s->dsp.diff_pixels(temp, src1, src2, stride);
882 s->block_last_index[0 /* FIXME */] =
884 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
890 length = s->intra_ac_vlc_length;
891 last_length = s->intra_ac_vlc_last_length;
892 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
895 length = s->inter_ac_vlc_length;
896 last_length = s->inter_ac_vlc_last_length;
899 if (last >= start_i) {
901 for (i = start_i; i < last; i++) {
902 int j = scantable[i];
907 if ((level & (~127)) == 0)
908 bits += length[UNI_AC_ENC_INDEX(run, level)];
917 level = temp[i] + 64;
919 av_assert2(level - 64);
921 if ((level & (~127)) == 0)
922 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
930 #define VSAD_INTRA(size) \
931 static int vsad_intra ## size ## _c(MpegEncContext *c, \
932 uint8_t *s, uint8_t *dummy, \
935 int score = 0, x, y; \
937 for (y = 1; y < h; y++) { \
938 for (x = 0; x < size; x += 4) { \
939 score += FFABS(s[x] - s[x + stride]) + \
940 FFABS(s[x + 1] - s[x + stride + 1]) + \
941 FFABS(s[x + 2] - s[x + 2 + stride]) + \
942 FFABS(s[x + 3] - s[x + 3 + stride]); \
953 static int vsad ## size ## _c(MpegEncContext *c, \
954 uint8_t *s1, uint8_t *s2, \
957 int score = 0, x, y; \
959 for (y = 1; y < h; y++) { \
960 for (x = 0; x < size; x++) \
961 score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]); \
971 #define SQ(a) ((a) * (a))
972 #define VSSE_INTRA(size) \
973 static int vsse_intra ## size ## _c(MpegEncContext *c, \
974 uint8_t *s, uint8_t *dummy, \
977 int score = 0, x, y; \
979 for (y = 1; y < h; y++) { \
980 for (x = 0; x < size; x += 4) { \
981 score += SQ(s[x] - s[x + stride]) + \
982 SQ(s[x + 1] - s[x + stride + 1]) + \
983 SQ(s[x + 2] - s[x + stride + 2]) + \
984 SQ(s[x + 3] - s[x + stride + 3]); \
995 static int vsse ## size ## _c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, \
998 int score = 0, x, y; \
1000 for (y = 1; y < h; y++) { \
1001 for (x = 0; x < size; x++) \
1002 score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]); \
1012 #define WRAPPER8_16_SQ(name8, name16) \
1013 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src, \
1014 int stride, int h) \
1018 score += name8(s, dst, src, stride, 8); \
1019 score += name8(s, dst + 8, src + 8, stride, 8); \
1021 dst += 8 * stride; \
1022 src += 8 * stride; \
1023 score += name8(s, dst, src, stride, 8); \
1024 score += name8(s, dst + 8, src + 8, stride, 8); \
1029 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)
1030 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)
1031 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)
1033 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)
1035 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)
1036 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
1037 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
1038 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
1040 /* draw the edges of width 'w' of an image of size width, height */
1041 // FIXME: Check that this is OK for MPEG-4 interlaced.
1042 static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,
1043 int w, int h, int sides)
1045 uint8_t *ptr = buf, *last_line;
1048 /* left and right */
1049 for (i = 0; i < height; i++) {
1050 memset(ptr - w, ptr[0], w);
1051 memset(ptr + width, ptr[width - 1], w);
1055 /* top and bottom + corners */
1057 last_line = buf + (height - 1) * wrap;
1058 if (sides & EDGE_TOP)
1059 for (i = 0; i < h; i++)
1061 memcpy(buf - (i + 1) * wrap, buf, width + w + w);
1062 if (sides & EDGE_BOTTOM)
1063 for (i = 0; i < h; i++)
1065 memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);
1068 /* init static data */
1069 av_cold void ff_dsputil_static_init(void)
1073 for (i = 0; i < 512; i++)
1074 ff_square_tab[i] = (i - 256) * (i - 256);
1077 int ff_check_alignment(void)
1079 static int did_fail = 0;
1080 LOCAL_ALIGNED_16(int, aligned, [4]);
1082 if ((intptr_t)aligned & 15) {
1084 #if HAVE_MMX || HAVE_ALTIVEC
1085 av_log(NULL, AV_LOG_ERROR,
1086 "Compiler did not align stack variables. Libavcodec has been miscompiled\n"
1087 "and may be very slow or crash. This is not a bug in libavcodec,\n"
1088 "but in the compiler. You may try recompiling using gcc >= 4.2.\n"
1089 "Do not report crashes to FFmpeg developers.\n");
1098 av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
1100 const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;
1102 ff_check_alignment();
1105 if (avctx->bits_per_raw_sample == 10) {
1106 c->fdct = ff_jpeg_fdct_islow_10;
1107 c->fdct248 = ff_fdct248_islow_10;
1109 if (avctx->dct_algo == FF_DCT_FASTINT) {
1110 c->fdct = ff_fdct_ifast;
1111 c->fdct248 = ff_fdct_ifast248;
1112 } else if (avctx->dct_algo == FF_DCT_FAAN) {
1113 c->fdct = ff_faandct;
1114 c->fdct248 = ff_faandct248;
1116 c->fdct = ff_jpeg_fdct_islow_8; // slow/accurate/default
1117 c->fdct248 = ff_fdct248_islow_8;
1120 #endif /* CONFIG_ENCODERS */
1122 c->diff_pixels = diff_pixels_c;
1124 c->sum_abs_dctelem = sum_abs_dctelem_c;
1126 c->pix_sum = pix_sum_c;
1127 c->pix_norm1 = pix_norm1_c;
1129 /* TODO [0] 16 [1] 8 */
1130 c->pix_abs[0][0] = pix_abs16_c;
1131 c->pix_abs[0][1] = pix_abs16_x2_c;
1132 c->pix_abs[0][2] = pix_abs16_y2_c;
1133 c->pix_abs[0][3] = pix_abs16_xy2_c;
1134 c->pix_abs[1][0] = pix_abs8_c;
1135 c->pix_abs[1][1] = pix_abs8_x2_c;
1136 c->pix_abs[1][2] = pix_abs8_y2_c;
1137 c->pix_abs[1][3] = pix_abs8_xy2_c;
1139 #define SET_CMP_FUNC(name) \
1140 c->name[0] = name ## 16_c; \
1141 c->name[1] = name ## 8x8_c;
1143 SET_CMP_FUNC(hadamard8_diff)
1144 c->hadamard8_diff[4] = hadamard8_intra16_c;
1145 c->hadamard8_diff[5] = hadamard8_intra8x8_c;
1146 SET_CMP_FUNC(dct_sad)
1147 SET_CMP_FUNC(dct_max)
1149 SET_CMP_FUNC(dct264_sad)
1151 c->sad[0] = pix_abs16_c;
1152 c->sad[1] = pix_abs8_c;
1153 c->sse[0] = sse16_c;
1156 SET_CMP_FUNC(quant_psnr)
1159 c->vsad[0] = vsad16_c;
1160 c->vsad[1] = vsad8_c;
1161 c->vsad[4] = vsad_intra16_c;
1162 c->vsad[5] = vsad_intra8_c;
1163 c->vsse[0] = vsse16_c;
1164 c->vsse[1] = vsse8_c;
1165 c->vsse[4] = vsse_intra16_c;
1166 c->vsse[5] = vsse_intra8_c;
1167 c->nsse[0] = nsse16_c;
1168 c->nsse[1] = nsse8_c;
1169 #if CONFIG_SNOW_DECODER || CONFIG_SNOW_ENCODER
1170 ff_dsputil_init_dwt(c);
1173 c->try_8x8basis = try_8x8basis_c;
1174 c->add_8x8basis = add_8x8basis_c;
1176 c->shrink[0] = av_image_copy_plane;
1177 c->shrink[1] = ff_shrink22;
1178 c->shrink[2] = ff_shrink44;
1179 c->shrink[3] = ff_shrink88;
1181 c->draw_edges = draw_edges_8_c;
1183 switch (avctx->bits_per_raw_sample) {
1188 c->get_pixels = get_pixels_16_c;
1191 if (avctx->bits_per_raw_sample<=8 || avctx->codec_type != AVMEDIA_TYPE_VIDEO) {
1192 c->get_pixels = get_pixels_8_c;
1199 ff_dsputil_init_alpha(c, avctx);
1201 ff_dsputil_init_arm(c, avctx, high_bit_depth);
1203 ff_dsputil_init_ppc(c, avctx, high_bit_depth);
1205 ff_dsputil_init_x86(c, avctx, high_bit_depth);
1208 av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
1210 ff_dsputil_init(c, avctx);
1213 av_cold void avpriv_dsputil_init(DSPContext *c, AVCodecContext *avctx)
1215 ff_dsputil_init(c, avctx);