3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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"
31 #include "copy_block.h"
34 #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 av_cold void ff_init_scantable(uint8_t *permutation, ScanTable *st,
52 const uint8_t *src_scantable)
56 st->scantable = src_scantable;
58 for (i = 0; i < 64; i++) {
59 int j = src_scantable[i];
60 st->permutated[i] = permutation[j];
64 for (i = 0; i < 64; i++) {
65 int j = st->permutated[i];
68 st->raster_end[i] = end;
72 av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,
73 int idct_permutation_type)
78 if (ff_init_scantable_permutation_x86(idct_permutation,
79 idct_permutation_type))
82 switch (idct_permutation_type) {
84 for (i = 0; i < 64; i++)
85 idct_permutation[i] = i;
87 case FF_LIBMPEG2_IDCT_PERM:
88 for (i = 0; i < 64; i++)
89 idct_permutation[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
91 case FF_TRANSPOSE_IDCT_PERM:
92 for (i = 0; i < 64; i++)
93 idct_permutation[i] = ((i & 7) << 3) | (i >> 3);
95 case FF_PARTTRANS_IDCT_PERM:
96 for (i = 0; i < 64; i++)
97 idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3);
100 av_log(NULL, AV_LOG_ERROR,
101 "Internal error, IDCT permutation not set\n");
105 static int pix_sum_c(uint8_t *pix, int line_size)
109 for (i = 0; i < 16; i++) {
110 for (j = 0; j < 16; j += 8) {
121 pix += line_size - 16;
126 static int pix_norm1_c(uint8_t *pix, int line_size)
129 uint32_t *sq = ff_square_tab + 256;
131 for (i = 0; i < 16; i++) {
132 for (j = 0; j < 16; j += 8) {
144 register uint64_t x = *(uint64_t *) pix;
146 s += sq[(x >> 8) & 0xff];
147 s += sq[(x >> 16) & 0xff];
148 s += sq[(x >> 24) & 0xff];
149 s += sq[(x >> 32) & 0xff];
150 s += sq[(x >> 40) & 0xff];
151 s += sq[(x >> 48) & 0xff];
152 s += sq[(x >> 56) & 0xff];
154 register uint32_t x = *(uint32_t *) pix;
156 s += sq[(x >> 8) & 0xff];
157 s += sq[(x >> 16) & 0xff];
158 s += sq[(x >> 24) & 0xff];
159 x = *(uint32_t *) (pix + 4);
161 s += sq[(x >> 8) & 0xff];
162 s += sq[(x >> 16) & 0xff];
163 s += sq[(x >> 24) & 0xff];
168 pix += line_size - 16;
173 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
174 int line_size, int h)
177 uint32_t *sq = ff_square_tab + 256;
179 for (i = 0; i < h; i++) {
180 s += sq[pix1[0] - pix2[0]];
181 s += sq[pix1[1] - pix2[1]];
182 s += sq[pix1[2] - pix2[2]];
183 s += sq[pix1[3] - pix2[3]];
190 static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
191 int line_size, int h)
194 uint32_t *sq = ff_square_tab + 256;
196 for (i = 0; i < h; i++) {
197 s += sq[pix1[0] - pix2[0]];
198 s += sq[pix1[1] - pix2[1]];
199 s += sq[pix1[2] - pix2[2]];
200 s += sq[pix1[3] - pix2[3]];
201 s += sq[pix1[4] - pix2[4]];
202 s += sq[pix1[5] - pix2[5]];
203 s += sq[pix1[6] - pix2[6]];
204 s += sq[pix1[7] - pix2[7]];
211 static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
212 int line_size, int h)
215 uint32_t *sq = ff_square_tab + 256;
217 for (i = 0; i < h; i++) {
218 s += sq[pix1[0] - pix2[0]];
219 s += sq[pix1[1] - pix2[1]];
220 s += sq[pix1[2] - pix2[2]];
221 s += sq[pix1[3] - pix2[3]];
222 s += sq[pix1[4] - pix2[4]];
223 s += sq[pix1[5] - pix2[5]];
224 s += sq[pix1[6] - pix2[6]];
225 s += sq[pix1[7] - pix2[7]];
226 s += sq[pix1[8] - pix2[8]];
227 s += sq[pix1[9] - pix2[9]];
228 s += sq[pix1[10] - pix2[10]];
229 s += sq[pix1[11] - pix2[11]];
230 s += sq[pix1[12] - pix2[12]];
231 s += sq[pix1[13] - pix2[13]];
232 s += sq[pix1[14] - pix2[14]];
233 s += sq[pix1[15] - pix2[15]];
241 static void diff_pixels_c(int16_t *restrict block, const uint8_t *s1,
242 const uint8_t *s2, int stride)
246 /* read the pixels */
247 for (i = 0; i < 8; i++) {
248 block[0] = s1[0] - s2[0];
249 block[1] = s1[1] - s2[1];
250 block[2] = s1[2] - s2[2];
251 block[3] = s1[3] - s2[3];
252 block[4] = s1[4] - s2[4];
253 block[5] = s1[5] - s2[5];
254 block[6] = s1[6] - s2[6];
255 block[7] = s1[7] - s2[7];
262 static void put_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,
267 /* read the pixels */
268 for (i = 0; i < 8; i++) {
269 pixels[0] = av_clip_uint8(block[0]);
270 pixels[1] = av_clip_uint8(block[1]);
271 pixels[2] = av_clip_uint8(block[2]);
272 pixels[3] = av_clip_uint8(block[3]);
273 pixels[4] = av_clip_uint8(block[4]);
274 pixels[5] = av_clip_uint8(block[5]);
275 pixels[6] = av_clip_uint8(block[6]);
276 pixels[7] = av_clip_uint8(block[7]);
283 static void put_signed_pixels_clamped_c(const int16_t *block,
284 uint8_t *restrict pixels,
289 for (i = 0; i < 8; i++) {
290 for (j = 0; j < 8; j++) {
293 else if (*block > 127)
296 *pixels = (uint8_t) (*block + 128);
300 pixels += (line_size - 8);
304 static void add_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,
309 /* read the pixels */
310 for (i = 0; i < 8; i++) {
311 pixels[0] = av_clip_uint8(pixels[0] + block[0]);
312 pixels[1] = av_clip_uint8(pixels[1] + block[1]);
313 pixels[2] = av_clip_uint8(pixels[2] + block[2]);
314 pixels[3] = av_clip_uint8(pixels[3] + block[3]);
315 pixels[4] = av_clip_uint8(pixels[4] + block[4]);
316 pixels[5] = av_clip_uint8(pixels[5] + block[5]);
317 pixels[6] = av_clip_uint8(pixels[6] + block[6]);
318 pixels[7] = av_clip_uint8(pixels[7] + block[7]);
324 static int sum_abs_dctelem_c(int16_t *block)
328 for (i = 0; i < 64; i++)
329 sum += FFABS(block[i]);
333 #define avg2(a, b) ((a + b + 1) >> 1)
334 #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)
336 static inline int pix_abs16_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]);
350 s += abs(pix1[8] - pix2[8]);
351 s += abs(pix1[9] - pix2[9]);
352 s += abs(pix1[10] - pix2[10]);
353 s += abs(pix1[11] - pix2[11]);
354 s += abs(pix1[12] - pix2[12]);
355 s += abs(pix1[13] - pix2[13]);
356 s += abs(pix1[14] - pix2[14]);
357 s += abs(pix1[15] - pix2[15]);
364 static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
365 int line_size, int h)
369 for (i = 0; i < h; i++) {
370 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
371 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
372 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
373 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
374 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
375 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
376 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
377 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
378 s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
379 s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
380 s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
381 s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
382 s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
383 s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
384 s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
385 s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
392 static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
393 int line_size, int h)
396 uint8_t *pix3 = pix2 + line_size;
398 for (i = 0; i < h; i++) {
399 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
400 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
401 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
402 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
403 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
404 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
405 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
406 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
407 s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
408 s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
409 s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
410 s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
411 s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
412 s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
413 s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
414 s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
422 static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
423 int line_size, int h)
426 uint8_t *pix3 = pix2 + line_size;
428 for (i = 0; i < h; i++) {
429 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
430 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
431 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
432 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
433 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
434 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
435 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
436 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
437 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
438 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
439 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
440 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
441 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
442 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
443 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
444 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
452 static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
453 int line_size, int h)
457 for (i = 0; i < h; i++) {
458 s += abs(pix1[0] - pix2[0]);
459 s += abs(pix1[1] - pix2[1]);
460 s += abs(pix1[2] - pix2[2]);
461 s += abs(pix1[3] - pix2[3]);
462 s += abs(pix1[4] - pix2[4]);
463 s += abs(pix1[5] - pix2[5]);
464 s += abs(pix1[6] - pix2[6]);
465 s += abs(pix1[7] - pix2[7]);
472 static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
473 int line_size, int h)
477 for (i = 0; i < h; i++) {
478 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
479 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
480 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
481 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
482 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
483 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
484 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
485 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
492 static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
493 int line_size, int h)
496 uint8_t *pix3 = pix2 + line_size;
498 for (i = 0; i < h; i++) {
499 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
500 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
501 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
502 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
503 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
504 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
505 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
506 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
514 static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
515 int line_size, int h)
518 uint8_t *pix3 = pix2 + line_size;
520 for (i = 0; i < h; i++) {
521 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
522 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
523 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
524 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
525 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
526 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
527 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
528 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
536 static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
538 int score1 = 0, score2 = 0, x, y;
540 for (y = 0; y < h; y++) {
541 for (x = 0; x < 16; x++)
542 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
544 for (x = 0; x < 15; x++)
545 score2 += FFABS(s1[x] - s1[x + stride] -
546 s1[x + 1] + s1[x + stride + 1]) -
547 FFABS(s2[x] - s2[x + stride] -
548 s2[x + 1] + s2[x + stride + 1]);
555 return score1 + FFABS(score2) * c->avctx->nsse_weight;
557 return score1 + FFABS(score2) * 8;
560 static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
562 int score1 = 0, score2 = 0, x, y;
564 for (y = 0; y < h; y++) {
565 for (x = 0; x < 8; x++)
566 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
568 for (x = 0; x < 7; x++)
569 score2 += FFABS(s1[x] - s1[x + stride] -
570 s1[x + 1] + s1[x + stride + 1]) -
571 FFABS(s2[x] - s2[x + stride] -
572 s2[x + 1] + s2[x + stride + 1]);
579 return score1 + FFABS(score2) * c->avctx->nsse_weight;
581 return score1 + FFABS(score2) * 8;
584 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],
585 int16_t basis[64], int scale)
588 unsigned int sum = 0;
590 for (i = 0; i < 8 * 8; i++) {
591 int b = rem[i] + ((basis[i] * scale +
592 (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>
593 (BASIS_SHIFT - RECON_SHIFT));
596 assert(-512 < b && b < 512);
598 sum += (w * b) * (w * b) >> 4;
603 static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)
607 for (i = 0; i < 8 * 8; i++)
608 rem[i] += (basis[i] * scale +
609 (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>
610 (BASIS_SHIFT - RECON_SHIFT);
613 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
619 void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)
623 memset(cmp, 0, sizeof(void *) * 6);
625 for (i = 0; i < 6; i++) {
626 switch (type & 0xFF) {
631 cmp[i] = c->hadamard8_diff[i];
637 cmp[i] = c->dct_sad[i];
640 cmp[i] = c->dct264_sad[i];
643 cmp[i] = c->dct_max[i];
646 cmp[i] = c->quant_psnr[i];
667 av_log(NULL, AV_LOG_ERROR,
668 "internal error in cmp function selection\n");
673 #define BUTTERFLY2(o1, o2, i1, i2) \
677 #define BUTTERFLY1(x, y) \
686 #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))
688 static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,
689 uint8_t *src, int stride, int h)
691 int i, temp[64], sum = 0;
695 for (i = 0; i < 8; i++) {
696 // FIXME: try pointer walks
697 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
698 src[stride * i + 0] - dst[stride * i + 0],
699 src[stride * i + 1] - dst[stride * i + 1]);
700 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
701 src[stride * i + 2] - dst[stride * i + 2],
702 src[stride * i + 3] - dst[stride * i + 3]);
703 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
704 src[stride * i + 4] - dst[stride * i + 4],
705 src[stride * i + 5] - dst[stride * i + 5]);
706 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
707 src[stride * i + 6] - dst[stride * i + 6],
708 src[stride * i + 7] - dst[stride * i + 7]);
710 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
711 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
712 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
713 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
715 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
716 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
717 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
718 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
721 for (i = 0; i < 8; i++) {
722 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
723 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
724 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
725 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
727 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
728 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
729 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
730 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
732 sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +
733 BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +
734 BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +
735 BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
740 static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,
741 uint8_t *dummy, int stride, int h)
743 int i, temp[64], sum = 0;
747 for (i = 0; i < 8; i++) {
748 // FIXME: try pointer walks
749 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
750 src[stride * i + 0], src[stride * i + 1]);
751 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
752 src[stride * i + 2], src[stride * i + 3]);
753 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
754 src[stride * i + 4], src[stride * i + 5]);
755 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
756 src[stride * i + 6], src[stride * i + 7]);
758 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
759 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
760 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
761 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
763 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
764 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
765 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
766 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
769 for (i = 0; i < 8; i++) {
770 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
771 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
772 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
773 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
775 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
776 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
777 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
778 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
781 BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])
782 + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])
783 + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])
784 + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
787 sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean
792 static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
793 uint8_t *src2, int stride, int h)
795 LOCAL_ALIGNED_16(int16_t, temp, [64]);
799 s->dsp.diff_pixels(temp, src1, src2, stride);
801 return s->dsp.sum_abs_dctelem(temp);
807 const int s07 = SRC(0) + SRC(7); \
808 const int s16 = SRC(1) + SRC(6); \
809 const int s25 = SRC(2) + SRC(5); \
810 const int s34 = SRC(3) + SRC(4); \
811 const int a0 = s07 + s34; \
812 const int a1 = s16 + s25; \
813 const int a2 = s07 - s34; \
814 const int a3 = s16 - s25; \
815 const int d07 = SRC(0) - SRC(7); \
816 const int d16 = SRC(1) - SRC(6); \
817 const int d25 = SRC(2) - SRC(5); \
818 const int d34 = SRC(3) - SRC(4); \
819 const int a4 = d16 + d25 + (d07 + (d07 >> 1)); \
820 const int a5 = d07 - d34 - (d25 + (d25 >> 1)); \
821 const int a6 = d07 + d34 - (d16 + (d16 >> 1)); \
822 const int a7 = d16 - d25 + (d34 + (d34 >> 1)); \
824 DST(1, a4 + (a7 >> 2)); \
825 DST(2, a2 + (a3 >> 1)); \
826 DST(3, a5 + (a6 >> 2)); \
828 DST(5, a6 - (a5 >> 2)); \
829 DST(6, (a2 >> 1) - a3); \
830 DST(7, (a4 >> 2) - a7); \
833 static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
834 uint8_t *src2, int stride, int h)
839 s->dsp.diff_pixels(dct[0], src1, src2, stride);
841 #define SRC(x) dct[i][x]
842 #define DST(x, v) dct[i][x] = v
843 for (i = 0; i < 8; i++)
848 #define SRC(x) dct[x][i]
849 #define DST(x, v) sum += FFABS(v)
850 for (i = 0; i < 8; i++)
858 static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
859 uint8_t *src2, int stride, int h)
861 LOCAL_ALIGNED_16(int16_t, temp, [64]);
866 s->dsp.diff_pixels(temp, src1, src2, stride);
869 for (i = 0; i < 64; i++)
870 sum = FFMAX(sum, FFABS(temp[i]));
875 static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
876 uint8_t *src2, int stride, int h)
878 LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);
879 int16_t *const bak = temp + 64;
885 s->dsp.diff_pixels(temp, src1, src2, stride);
887 memcpy(bak, temp, 64 * sizeof(int16_t));
889 s->block_last_index[0 /* FIXME */] =
890 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
891 s->dct_unquantize_inter(s, temp, 0, s->qscale);
892 ff_simple_idct_8(temp); // FIXME
894 for (i = 0; i < 64; i++)
895 sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);
900 static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
903 const uint8_t *scantable = s->intra_scantable.permutated;
904 LOCAL_ALIGNED_16(int16_t, temp, [64]);
905 LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);
906 LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);
907 int i, last, run, bits, level, distortion, start_i;
908 const int esc_length = s->ac_esc_length;
909 uint8_t *length, *last_length;
913 copy_block8(lsrc1, src1, 8, stride, 8);
914 copy_block8(lsrc2, src2, 8, stride, 8);
916 s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);
918 s->block_last_index[0 /* FIXME */] =
920 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
926 length = s->intra_ac_vlc_length;
927 last_length = s->intra_ac_vlc_last_length;
928 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
931 length = s->inter_ac_vlc_length;
932 last_length = s->inter_ac_vlc_last_length;
935 if (last >= start_i) {
937 for (i = start_i; i < last; i++) {
938 int j = scantable[i];
943 if ((level & (~127)) == 0)
944 bits += length[UNI_AC_ENC_INDEX(run, level)];
953 level = temp[i] + 64;
957 if ((level & (~127)) == 0) {
958 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
965 s->dct_unquantize_intra(s, temp, 0, s->qscale);
967 s->dct_unquantize_inter(s, temp, 0, s->qscale);
970 s->dsp.idct_add(lsrc2, 8, temp);
972 distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);
974 return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);
977 static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
980 const uint8_t *scantable = s->intra_scantable.permutated;
981 LOCAL_ALIGNED_16(int16_t, temp, [64]);
982 int i, last, run, bits, level, start_i;
983 const int esc_length = s->ac_esc_length;
984 uint8_t *length, *last_length;
988 s->dsp.diff_pixels(temp, src1, src2, stride);
990 s->block_last_index[0 /* FIXME */] =
992 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
998 length = s->intra_ac_vlc_length;
999 last_length = s->intra_ac_vlc_last_length;
1000 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
1003 length = s->inter_ac_vlc_length;
1004 last_length = s->inter_ac_vlc_last_length;
1007 if (last >= start_i) {
1009 for (i = start_i; i < last; i++) {
1010 int j = scantable[i];
1015 if ((level & (~127)) == 0)
1016 bits += length[UNI_AC_ENC_INDEX(run, level)];
1023 i = scantable[last];
1025 level = temp[i] + 64;
1029 if ((level & (~127)) == 0)
1030 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
1038 #define VSAD_INTRA(size) \
1039 static int vsad_intra ## size ## _c(MpegEncContext *c, \
1040 uint8_t *s, uint8_t *dummy, \
1041 int stride, int h) \
1043 int score = 0, x, y; \
1045 for (y = 1; y < h; y++) { \
1046 for (x = 0; x < size; x += 4) { \
1047 score += FFABS(s[x] - s[x + stride]) + \
1048 FFABS(s[x + 1] - s[x + stride + 1]) + \
1049 FFABS(s[x + 2] - s[x + 2 + stride]) + \
1050 FFABS(s[x + 3] - s[x + 3 + stride]); \
1060 static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
1063 int score = 0, x, y;
1065 for (y = 1; y < h; y++) {
1066 for (x = 0; x < 16; x++)
1067 score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);
1075 #define SQ(a) ((a) * (a))
1076 #define VSSE_INTRA(size) \
1077 static int vsse_intra ## size ## _c(MpegEncContext *c, \
1078 uint8_t *s, uint8_t *dummy, \
1079 int stride, int h) \
1081 int score = 0, x, y; \
1083 for (y = 1; y < h; y++) { \
1084 for (x = 0; x < size; x += 4) { \
1085 score += SQ(s[x] - s[x + stride]) + \
1086 SQ(s[x + 1] - s[x + stride + 1]) + \
1087 SQ(s[x + 2] - s[x + stride + 2]) + \
1088 SQ(s[x + 3] - s[x + stride + 3]); \
1098 static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
1101 int score = 0, x, y;
1103 for (y = 1; y < h; y++) {
1104 for (x = 0; x < 16; x++)
1105 score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);
1113 #define WRAPPER8_16_SQ(name8, name16) \
1114 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src, \
1115 int stride, int h) \
1119 score += name8(s, dst, src, stride, 8); \
1120 score += name8(s, dst + 8, src + 8, stride, 8); \
1122 dst += 8 * stride; \
1123 src += 8 * stride; \
1124 score += name8(s, dst, src, stride, 8); \
1125 score += name8(s, dst + 8, src + 8, stride, 8); \
1130 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)
1131 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)
1132 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)
1134 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)
1136 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)
1137 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
1138 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
1139 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
1141 static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)
1143 ff_j_rev_dct(block);
1144 put_pixels_clamped_c(block, dest, line_size);
1147 static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)
1149 ff_j_rev_dct(block);
1150 add_pixels_clamped_c(block, dest, line_size);
1153 /* draw the edges of width 'w' of an image of size width, height */
1154 // FIXME: Check that this is OK for MPEG-4 interlaced.
1155 static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,
1156 int w, int h, int sides)
1158 uint8_t *ptr = buf, *last_line;
1161 /* left and right */
1162 for (i = 0; i < height; i++) {
1163 memset(ptr - w, ptr[0], w);
1164 memset(ptr + width, ptr[width - 1], w);
1168 /* top and bottom + corners */
1170 last_line = buf + (height - 1) * wrap;
1171 if (sides & EDGE_TOP)
1172 for (i = 0; i < h; i++)
1174 memcpy(buf - (i + 1) * wrap, buf, width + w + w);
1175 if (sides & EDGE_BOTTOM)
1176 for (i = 0; i < h; i++)
1178 memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);
1181 /* init static data */
1182 av_cold void ff_dsputil_static_init(void)
1186 for (i = 0; i < 512; i++)
1187 ff_square_tab[i] = (i - 256) * (i - 256);
1190 av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
1192 const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;
1195 if (avctx->bits_per_raw_sample == 10) {
1196 c->fdct = ff_jpeg_fdct_islow_10;
1197 c->fdct248 = ff_fdct248_islow_10;
1199 if (avctx->dct_algo == FF_DCT_FASTINT) {
1200 c->fdct = ff_fdct_ifast;
1201 c->fdct248 = ff_fdct_ifast248;
1202 } else if (avctx->dct_algo == FF_DCT_FAAN) {
1203 c->fdct = ff_faandct;
1204 c->fdct248 = ff_faandct248;
1206 c->fdct = ff_jpeg_fdct_islow_8; // slow/accurate/default
1207 c->fdct248 = ff_fdct248_islow_8;
1210 #endif /* CONFIG_ENCODERS */
1212 if (avctx->bits_per_raw_sample == 10) {
1213 c->idct_put = ff_simple_idct_put_10;
1214 c->idct_add = ff_simple_idct_add_10;
1215 c->idct = ff_simple_idct_10;
1216 c->idct_permutation_type = FF_NO_IDCT_PERM;
1218 if (avctx->idct_algo == FF_IDCT_INT) {
1219 c->idct_put = jref_idct_put;
1220 c->idct_add = jref_idct_add;
1221 c->idct = ff_j_rev_dct;
1222 c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;
1223 } else if (avctx->idct_algo == FF_IDCT_FAAN) {
1224 c->idct_put = ff_faanidct_put;
1225 c->idct_add = ff_faanidct_add;
1226 c->idct = ff_faanidct;
1227 c->idct_permutation_type = FF_NO_IDCT_PERM;
1228 } else { // accurate/default
1229 c->idct_put = ff_simple_idct_put_8;
1230 c->idct_add = ff_simple_idct_add_8;
1231 c->idct = ff_simple_idct_8;
1232 c->idct_permutation_type = FF_NO_IDCT_PERM;
1236 c->diff_pixels = diff_pixels_c;
1238 c->put_pixels_clamped = put_pixels_clamped_c;
1239 c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;
1240 c->add_pixels_clamped = add_pixels_clamped_c;
1242 c->sum_abs_dctelem = sum_abs_dctelem_c;
1244 c->pix_sum = pix_sum_c;
1245 c->pix_norm1 = pix_norm1_c;
1247 /* TODO [0] 16 [1] 8 */
1248 c->pix_abs[0][0] = pix_abs16_c;
1249 c->pix_abs[0][1] = pix_abs16_x2_c;
1250 c->pix_abs[0][2] = pix_abs16_y2_c;
1251 c->pix_abs[0][3] = pix_abs16_xy2_c;
1252 c->pix_abs[1][0] = pix_abs8_c;
1253 c->pix_abs[1][1] = pix_abs8_x2_c;
1254 c->pix_abs[1][2] = pix_abs8_y2_c;
1255 c->pix_abs[1][3] = pix_abs8_xy2_c;
1257 #define SET_CMP_FUNC(name) \
1258 c->name[0] = name ## 16_c; \
1259 c->name[1] = name ## 8x8_c;
1261 SET_CMP_FUNC(hadamard8_diff)
1262 c->hadamard8_diff[4] = hadamard8_intra16_c;
1263 c->hadamard8_diff[5] = hadamard8_intra8x8_c;
1264 SET_CMP_FUNC(dct_sad)
1265 SET_CMP_FUNC(dct_max)
1267 SET_CMP_FUNC(dct264_sad)
1269 c->sad[0] = pix_abs16_c;
1270 c->sad[1] = pix_abs8_c;
1271 c->sse[0] = sse16_c;
1274 SET_CMP_FUNC(quant_psnr)
1277 c->vsad[0] = vsad16_c;
1278 c->vsad[4] = vsad_intra16_c;
1279 c->vsad[5] = vsad_intra8_c;
1280 c->vsse[0] = vsse16_c;
1281 c->vsse[4] = vsse_intra16_c;
1282 c->vsse[5] = vsse_intra8_c;
1283 c->nsse[0] = nsse16_c;
1284 c->nsse[1] = nsse8_c;
1286 c->try_8x8basis = try_8x8basis_c;
1287 c->add_8x8basis = add_8x8basis_c;
1289 c->shrink[0] = av_image_copy_plane;
1290 c->shrink[1] = ff_shrink22;
1291 c->shrink[2] = ff_shrink44;
1292 c->shrink[3] = ff_shrink88;
1294 c->draw_edges = draw_edges_8_c;
1296 switch (avctx->bits_per_raw_sample) {
1299 c->get_pixels = get_pixels_16_c;
1302 c->get_pixels = get_pixels_8_c;
1307 ff_dsputil_init_arm(c, avctx, high_bit_depth);
1309 ff_dsputil_init_ppc(c, avctx, high_bit_depth);
1311 ff_dsputil_init_x86(c, avctx, high_bit_depth);
1313 ff_init_scantable_permutation(c->idct_permutation,
1314 c->idct_permutation_type);