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
23 #include "libavutil/attributes.h"
24 #include "libavutil/internal.h"
26 #include "copy_block.h"
27 #include "simple_idct.h"
29 #include "mpegvideo.h"
32 /* (i - 256) * (i - 256) */
33 const uint32_t ff_square_tab[512] = {
34 65536, 65025, 64516, 64009, 63504, 63001, 62500, 62001, 61504, 61009, 60516, 60025, 59536, 59049, 58564, 58081,
35 57600, 57121, 56644, 56169, 55696, 55225, 54756, 54289, 53824, 53361, 52900, 52441, 51984, 51529, 51076, 50625,
36 50176, 49729, 49284, 48841, 48400, 47961, 47524, 47089, 46656, 46225, 45796, 45369, 44944, 44521, 44100, 43681,
37 43264, 42849, 42436, 42025, 41616, 41209, 40804, 40401, 40000, 39601, 39204, 38809, 38416, 38025, 37636, 37249,
38 36864, 36481, 36100, 35721, 35344, 34969, 34596, 34225, 33856, 33489, 33124, 32761, 32400, 32041, 31684, 31329,
39 30976, 30625, 30276, 29929, 29584, 29241, 28900, 28561, 28224, 27889, 27556, 27225, 26896, 26569, 26244, 25921,
40 25600, 25281, 24964, 24649, 24336, 24025, 23716, 23409, 23104, 22801, 22500, 22201, 21904, 21609, 21316, 21025,
41 20736, 20449, 20164, 19881, 19600, 19321, 19044, 18769, 18496, 18225, 17956, 17689, 17424, 17161, 16900, 16641,
42 16384, 16129, 15876, 15625, 15376, 15129, 14884, 14641, 14400, 14161, 13924, 13689, 13456, 13225, 12996, 12769,
43 12544, 12321, 12100, 11881, 11664, 11449, 11236, 11025, 10816, 10609, 10404, 10201, 10000, 9801, 9604, 9409,
44 9216, 9025, 8836, 8649, 8464, 8281, 8100, 7921, 7744, 7569, 7396, 7225, 7056, 6889, 6724, 6561,
45 6400, 6241, 6084, 5929, 5776, 5625, 5476, 5329, 5184, 5041, 4900, 4761, 4624, 4489, 4356, 4225,
46 4096, 3969, 3844, 3721, 3600, 3481, 3364, 3249, 3136, 3025, 2916, 2809, 2704, 2601, 2500, 2401,
47 2304, 2209, 2116, 2025, 1936, 1849, 1764, 1681, 1600, 1521, 1444, 1369, 1296, 1225, 1156, 1089,
48 1024, 961, 900, 841, 784, 729, 676, 625, 576, 529, 484, 441, 400, 361, 324, 289,
49 256, 225, 196, 169, 144, 121, 100, 81, 64, 49, 36, 25, 16, 9, 4, 1,
50 0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100, 121, 144, 169, 196, 225,
51 256, 289, 324, 361, 400, 441, 484, 529, 576, 625, 676, 729, 784, 841, 900, 961,
52 1024, 1089, 1156, 1225, 1296, 1369, 1444, 1521, 1600, 1681, 1764, 1849, 1936, 2025, 2116, 2209,
53 2304, 2401, 2500, 2601, 2704, 2809, 2916, 3025, 3136, 3249, 3364, 3481, 3600, 3721, 3844, 3969,
54 4096, 4225, 4356, 4489, 4624, 4761, 4900, 5041, 5184, 5329, 5476, 5625, 5776, 5929, 6084, 6241,
55 6400, 6561, 6724, 6889, 7056, 7225, 7396, 7569, 7744, 7921, 8100, 8281, 8464, 8649, 8836, 9025,
56 9216, 9409, 9604, 9801, 10000, 10201, 10404, 10609, 10816, 11025, 11236, 11449, 11664, 11881, 12100, 12321,
57 12544, 12769, 12996, 13225, 13456, 13689, 13924, 14161, 14400, 14641, 14884, 15129, 15376, 15625, 15876, 16129,
58 16384, 16641, 16900, 17161, 17424, 17689, 17956, 18225, 18496, 18769, 19044, 19321, 19600, 19881, 20164, 20449,
59 20736, 21025, 21316, 21609, 21904, 22201, 22500, 22801, 23104, 23409, 23716, 24025, 24336, 24649, 24964, 25281,
60 25600, 25921, 26244, 26569, 26896, 27225, 27556, 27889, 28224, 28561, 28900, 29241, 29584, 29929, 30276, 30625,
61 30976, 31329, 31684, 32041, 32400, 32761, 33124, 33489, 33856, 34225, 34596, 34969, 35344, 35721, 36100, 36481,
62 36864, 37249, 37636, 38025, 38416, 38809, 39204, 39601, 40000, 40401, 40804, 41209, 41616, 42025, 42436, 42849,
63 43264, 43681, 44100, 44521, 44944, 45369, 45796, 46225, 46656, 47089, 47524, 47961, 48400, 48841, 49284, 49729,
64 50176, 50625, 51076, 51529, 51984, 52441, 52900, 53361, 53824, 54289, 54756, 55225, 55696, 56169, 56644, 57121,
65 57600, 58081, 58564, 59049, 59536, 60025, 60516, 61009, 61504, 62001, 62500, 63001, 63504, 64009, 64516, 65025,
68 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
69 ptrdiff_t stride, int h)
72 const uint32_t *sq = ff_square_tab + 256;
74 for (i = 0; i < h; i++) {
75 s += sq[pix1[0] - pix2[0]];
76 s += sq[pix1[1] - pix2[1]];
77 s += sq[pix1[2] - pix2[2]];
78 s += sq[pix1[3] - pix2[3]];
85 static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
86 ptrdiff_t stride, int h)
89 const uint32_t *sq = ff_square_tab + 256;
91 for (i = 0; i < h; i++) {
92 s += sq[pix1[0] - pix2[0]];
93 s += sq[pix1[1] - pix2[1]];
94 s += sq[pix1[2] - pix2[2]];
95 s += sq[pix1[3] - pix2[3]];
96 s += sq[pix1[4] - pix2[4]];
97 s += sq[pix1[5] - pix2[5]];
98 s += sq[pix1[6] - pix2[6]];
99 s += sq[pix1[7] - pix2[7]];
106 static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
107 ptrdiff_t stride, int h)
110 const uint32_t *sq = ff_square_tab + 256;
112 for (i = 0; i < h; i++) {
113 s += sq[pix1[0] - pix2[0]];
114 s += sq[pix1[1] - pix2[1]];
115 s += sq[pix1[2] - pix2[2]];
116 s += sq[pix1[3] - pix2[3]];
117 s += sq[pix1[4] - pix2[4]];
118 s += sq[pix1[5] - pix2[5]];
119 s += sq[pix1[6] - pix2[6]];
120 s += sq[pix1[7] - pix2[7]];
121 s += sq[pix1[8] - pix2[8]];
122 s += sq[pix1[9] - pix2[9]];
123 s += sq[pix1[10] - pix2[10]];
124 s += sq[pix1[11] - pix2[11]];
125 s += sq[pix1[12] - pix2[12]];
126 s += sq[pix1[13] - pix2[13]];
127 s += sq[pix1[14] - pix2[14]];
128 s += sq[pix1[15] - pix2[15]];
136 static int sum_abs_dctelem_c(int16_t *block)
140 for (i = 0; i < 64; i++)
141 sum += FFABS(block[i]);
145 #define avg2(a, b) (((a) + (b) + 1) >> 1)
146 #define avg4(a, b, c, d) (((a) + (b) + (c) + (d) + 2) >> 2)
148 static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
149 ptrdiff_t stride, int h)
153 for (i = 0; i < h; i++) {
154 s += abs(pix1[0] - pix2[0]);
155 s += abs(pix1[1] - pix2[1]);
156 s += abs(pix1[2] - pix2[2]);
157 s += abs(pix1[3] - pix2[3]);
158 s += abs(pix1[4] - pix2[4]);
159 s += abs(pix1[5] - pix2[5]);
160 s += abs(pix1[6] - pix2[6]);
161 s += abs(pix1[7] - pix2[7]);
162 s += abs(pix1[8] - pix2[8]);
163 s += abs(pix1[9] - pix2[9]);
164 s += abs(pix1[10] - pix2[10]);
165 s += abs(pix1[11] - pix2[11]);
166 s += abs(pix1[12] - pix2[12]);
167 s += abs(pix1[13] - pix2[13]);
168 s += abs(pix1[14] - pix2[14]);
169 s += abs(pix1[15] - pix2[15]);
176 static inline int pix_median_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
177 ptrdiff_t stride, int h)
181 #define V(x) (pix1[x] - pix2[x])
184 s += abs(V(1) - V(0));
185 s += abs(V(2) - V(1));
186 s += abs(V(3) - V(2));
187 s += abs(V(4) - V(3));
188 s += abs(V(5) - V(4));
189 s += abs(V(6) - V(5));
190 s += abs(V(7) - V(6));
191 s += abs(V(8) - V(7));
192 s += abs(V(9) - V(8));
193 s += abs(V(10) - V(9));
194 s += abs(V(11) - V(10));
195 s += abs(V(12) - V(11));
196 s += abs(V(13) - V(12));
197 s += abs(V(14) - V(13));
198 s += abs(V(15) - V(14));
203 for (i = 1; i < h; i++) {
204 s += abs(V(0) - V(-stride));
205 for (j = 1; j < 16; j++)
206 s += abs(V(j) - mid_pred(V(j-stride), V(j-1), V(j-stride) + V(j-1) - V(j-stride-1)));
215 static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
216 ptrdiff_t stride, int h)
220 for (i = 0; i < h; i++) {
221 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
222 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
223 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
224 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
225 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
226 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
227 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
228 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
229 s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
230 s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
231 s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
232 s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
233 s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
234 s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
235 s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
236 s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
243 static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
244 ptrdiff_t stride, int h)
247 uint8_t *pix3 = pix2 + stride;
249 for (i = 0; i < h; i++) {
250 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
251 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
252 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
253 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
254 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
255 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
256 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
257 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
258 s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
259 s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
260 s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
261 s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
262 s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
263 s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
264 s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
265 s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
273 static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
274 ptrdiff_t stride, int h)
277 uint8_t *pix3 = pix2 + stride;
279 for (i = 0; i < h; i++) {
280 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
281 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
282 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
283 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
284 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
285 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
286 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
287 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
288 s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
289 s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
290 s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
291 s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
292 s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
293 s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
294 s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
295 s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
303 static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
304 ptrdiff_t stride, int h)
308 for (i = 0; i < h; i++) {
309 s += abs(pix1[0] - pix2[0]);
310 s += abs(pix1[1] - pix2[1]);
311 s += abs(pix1[2] - pix2[2]);
312 s += abs(pix1[3] - pix2[3]);
313 s += abs(pix1[4] - pix2[4]);
314 s += abs(pix1[5] - pix2[5]);
315 s += abs(pix1[6] - pix2[6]);
316 s += abs(pix1[7] - pix2[7]);
323 static inline int pix_median_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
324 ptrdiff_t stride, int h)
328 #define V(x) (pix1[x] - pix2[x])
331 s += abs(V(1) - V(0));
332 s += abs(V(2) - V(1));
333 s += abs(V(3) - V(2));
334 s += abs(V(4) - V(3));
335 s += abs(V(5) - V(4));
336 s += abs(V(6) - V(5));
337 s += abs(V(7) - V(6));
342 for (i = 1; i < h; i++) {
343 s += abs(V(0) - V(-stride));
344 for (j = 1; j < 8; j++)
345 s += abs(V(j) - mid_pred(V(j-stride), V(j-1), V(j-stride) + V(j-1) - V(j-stride-1)));
354 static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
355 ptrdiff_t stride, int h)
359 for (i = 0; i < h; i++) {
360 s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
361 s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
362 s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
363 s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
364 s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
365 s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
366 s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
367 s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
374 static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
375 ptrdiff_t stride, int h)
378 uint8_t *pix3 = pix2 + stride;
380 for (i = 0; i < h; i++) {
381 s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
382 s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
383 s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
384 s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
385 s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
386 s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
387 s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
388 s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
396 static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
397 ptrdiff_t stride, int h)
400 uint8_t *pix3 = pix2 + stride;
402 for (i = 0; i < h; i++) {
403 s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
404 s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
405 s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
406 s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
407 s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
408 s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
409 s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
410 s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
418 static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
419 ptrdiff_t stride, int h)
421 int score1 = 0, score2 = 0, x, y;
423 for (y = 0; y < h; y++) {
424 for (x = 0; x < 16; x++)
425 score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
427 for (x = 0; x < 15; x++)
428 score2 += FFABS(s1[x] - s1[x + stride] -
429 s1[x + 1] + s1[x + stride + 1]) -
430 FFABS(s2[x] - s2[x + stride] -
431 s2[x + 1] + s2[x + stride + 1]);
438 return score1 + FFABS(score2) * c->avctx->nsse_weight;
440 return score1 + FFABS(score2) * 8;
443 static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,
444 ptrdiff_t 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 zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
469 ptrdiff_t stride, int h)
474 void ff_set_cmp(MECmpContext *c, me_cmp_func *cmp, int type)
478 memset(cmp, 0, sizeof(void *) * 6);
480 for (i = 0; i < 6; i++) {
481 switch (type & 0xFF) {
485 case FF_CMP_MEDIAN_SAD:
486 cmp[i] = c->median_sad[i];
489 cmp[i] = c->hadamard8_diff[i];
495 cmp[i] = c->dct_sad[i];
498 cmp[i] = c->dct264_sad[i];
501 cmp[i] = c->dct_max[i];
504 cmp[i] = c->quant_psnr[i];
533 av_log(NULL, AV_LOG_ERROR,
534 "internal error in cmp function selection\n");
539 #define BUTTERFLY2(o1, o2, i1, i2) \
543 #define BUTTERFLY1(x, y) \
552 #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))
554 static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,
555 uint8_t *src, ptrdiff_t stride, int h)
557 int i, temp[64], sum = 0;
561 for (i = 0; i < 8; i++) {
562 // FIXME: try pointer walks
563 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
564 src[stride * i + 0] - dst[stride * i + 0],
565 src[stride * i + 1] - dst[stride * i + 1]);
566 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
567 src[stride * i + 2] - dst[stride * i + 2],
568 src[stride * i + 3] - dst[stride * i + 3]);
569 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
570 src[stride * i + 4] - dst[stride * i + 4],
571 src[stride * i + 5] - dst[stride * i + 5]);
572 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
573 src[stride * i + 6] - dst[stride * i + 6],
574 src[stride * i + 7] - dst[stride * i + 7]);
576 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
577 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
578 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
579 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
581 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
582 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
583 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
584 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
587 for (i = 0; i < 8; i++) {
588 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
589 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
590 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
591 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
593 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
594 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
595 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
596 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
598 sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +
599 BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +
600 BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +
601 BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
606 static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,
607 uint8_t *dummy, ptrdiff_t stride, int h)
609 int i, temp[64], sum = 0;
613 for (i = 0; i < 8; i++) {
614 // FIXME: try pointer walks
615 BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
616 src[stride * i + 0], src[stride * i + 1]);
617 BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
618 src[stride * i + 2], src[stride * i + 3]);
619 BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
620 src[stride * i + 4], src[stride * i + 5]);
621 BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
622 src[stride * i + 6], src[stride * i + 7]);
624 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
625 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
626 BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
627 BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
629 BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
630 BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
631 BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
632 BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
635 for (i = 0; i < 8; i++) {
636 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
637 BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
638 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
639 BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
641 BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
642 BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
643 BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
644 BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
647 BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])
648 + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])
649 + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])
650 + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
653 sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean
658 static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
659 uint8_t *src2, ptrdiff_t stride, int h)
661 LOCAL_ALIGNED_16(int16_t, temp, [64]);
665 s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
667 return s->mecc.sum_abs_dctelem(temp);
673 const int s07 = SRC(0) + SRC(7); \
674 const int s16 = SRC(1) + SRC(6); \
675 const int s25 = SRC(2) + SRC(5); \
676 const int s34 = SRC(3) + SRC(4); \
677 const int a0 = s07 + s34; \
678 const int a1 = s16 + s25; \
679 const int a2 = s07 - s34; \
680 const int a3 = s16 - s25; \
681 const int d07 = SRC(0) - SRC(7); \
682 const int d16 = SRC(1) - SRC(6); \
683 const int d25 = SRC(2) - SRC(5); \
684 const int d34 = SRC(3) - SRC(4); \
685 const int a4 = d16 + d25 + (d07 + (d07 >> 1)); \
686 const int a5 = d07 - d34 - (d25 + (d25 >> 1)); \
687 const int a6 = d07 + d34 - (d16 + (d16 >> 1)); \
688 const int a7 = d16 - d25 + (d34 + (d34 >> 1)); \
690 DST(1, a4 + (a7 >> 2)); \
691 DST(2, a2 + (a3 >> 1)); \
692 DST(3, a5 + (a6 >> 2)); \
694 DST(5, a6 - (a5 >> 2)); \
695 DST(6, (a2 >> 1) - a3); \
696 DST(7, (a4 >> 2) - a7); \
699 static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
700 uint8_t *src2, ptrdiff_t stride, int h)
705 s->pdsp.diff_pixels_unaligned(dct[0], src1, src2, stride);
707 #define SRC(x) dct[i][x]
708 #define DST(x, v) dct[i][x] = v
709 for (i = 0; i < 8; i++)
714 #define SRC(x) dct[x][i]
715 #define DST(x, v) sum += FFABS(v)
716 for (i = 0; i < 8; i++)
724 static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
725 uint8_t *src2, ptrdiff_t stride, int h)
727 LOCAL_ALIGNED_16(int16_t, temp, [64]);
732 s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
735 for (i = 0; i < 64; i++)
736 sum = FFMAX(sum, FFABS(temp[i]));
741 static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
742 uint8_t *src2, ptrdiff_t stride, int h)
744 LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);
745 int16_t *const bak = temp + 64;
751 s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
753 memcpy(bak, temp, 64 * sizeof(int16_t));
755 s->block_last_index[0 /* FIXME */] =
756 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
757 s->dct_unquantize_inter(s, temp, 0, s->qscale);
758 ff_simple_idct_int16_8bit(temp); // FIXME
760 for (i = 0; i < 64; i++)
761 sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);
766 static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
767 ptrdiff_t stride, int h)
769 const uint8_t *scantable = s->intra_scantable.permutated;
770 LOCAL_ALIGNED_16(int16_t, temp, [64]);
771 LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);
772 LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);
773 int i, last, run, bits, level, distortion, start_i;
774 const int esc_length = s->ac_esc_length;
775 uint8_t *length, *last_length;
779 copy_block8(lsrc1, src1, 8, stride, 8);
780 copy_block8(lsrc2, src2, 8, stride, 8);
782 s->pdsp.diff_pixels(temp, lsrc1, lsrc2, 8);
784 s->block_last_index[0 /* FIXME */] =
786 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
792 length = s->intra_ac_vlc_length;
793 last_length = s->intra_ac_vlc_last_length;
794 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
797 length = s->inter_ac_vlc_length;
798 last_length = s->inter_ac_vlc_last_length;
801 if (last >= start_i) {
803 for (i = start_i; i < last; i++) {
804 int j = scantable[i];
809 if ((level & (~127)) == 0)
810 bits += length[UNI_AC_ENC_INDEX(run, level)];
819 level = temp[i] + 64;
821 av_assert2(level - 64);
823 if ((level & (~127)) == 0) {
824 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
831 s->dct_unquantize_intra(s, temp, 0, s->qscale);
833 s->dct_unquantize_inter(s, temp, 0, s->qscale);
836 s->idsp.idct_add(lsrc2, 8, temp);
838 distortion = s->mecc.sse[1](NULL, lsrc2, lsrc1, 8, 8);
840 return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);
843 static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
844 ptrdiff_t stride, int h)
846 const uint8_t *scantable = s->intra_scantable.permutated;
847 LOCAL_ALIGNED_16(int16_t, temp, [64]);
848 int i, last, run, bits, level, start_i;
849 const int esc_length = s->ac_esc_length;
850 uint8_t *length, *last_length;
854 s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
856 s->block_last_index[0 /* FIXME */] =
858 s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
864 length = s->intra_ac_vlc_length;
865 last_length = s->intra_ac_vlc_last_length;
866 bits += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
869 length = s->inter_ac_vlc_length;
870 last_length = s->inter_ac_vlc_last_length;
873 if (last >= start_i) {
875 for (i = start_i; i < last; i++) {
876 int j = scantable[i];
881 if ((level & (~127)) == 0)
882 bits += length[UNI_AC_ENC_INDEX(run, level)];
891 level = temp[i] + 64;
893 av_assert2(level - 64);
895 if ((level & (~127)) == 0)
896 bits += last_length[UNI_AC_ENC_INDEX(run, level)];
904 #define VSAD_INTRA(size) \
905 static int vsad_intra ## size ## _c(MpegEncContext *c, \
906 uint8_t *s, uint8_t *dummy, \
907 ptrdiff_t stride, int h) \
909 int score = 0, x, y; \
911 for (y = 1; y < h; y++) { \
912 for (x = 0; x < size; x += 4) { \
913 score += FFABS(s[x] - s[x + stride]) + \
914 FFABS(s[x + 1] - s[x + stride + 1]) + \
915 FFABS(s[x + 2] - s[x + 2 + stride]) + \
916 FFABS(s[x + 3] - s[x + 3 + stride]); \
927 static int vsad ## size ## _c(MpegEncContext *c, \
928 uint8_t *s1, uint8_t *s2, \
929 ptrdiff_t stride, int h) \
931 int score = 0, x, y; \
933 for (y = 1; y < h; y++) { \
934 for (x = 0; x < size; x++) \
935 score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]); \
945 #define SQ(a) ((a) * (a))
946 #define VSSE_INTRA(size) \
947 static int vsse_intra ## size ## _c(MpegEncContext *c, \
948 uint8_t *s, uint8_t *dummy, \
949 ptrdiff_t stride, int h) \
951 int score = 0, x, y; \
953 for (y = 1; y < h; y++) { \
954 for (x = 0; x < size; x += 4) { \
955 score += SQ(s[x] - s[x + stride]) + \
956 SQ(s[x + 1] - s[x + stride + 1]) + \
957 SQ(s[x + 2] - s[x + stride + 2]) + \
958 SQ(s[x + 3] - s[x + stride + 3]); \
969 static int vsse ## size ## _c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, \
970 ptrdiff_t stride, int h) \
972 int score = 0, x, y; \
974 for (y = 1; y < h; y++) { \
975 for (x = 0; x < size; x++) \
976 score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]); \
986 #define WRAPPER8_16_SQ(name8, name16) \
987 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src, \
988 ptrdiff_t stride, int h) \
992 score += name8(s, dst, src, stride, 8); \
993 score += name8(s, dst + 8, src + 8, stride, 8); \
997 score += name8(s, dst, src, stride, 8); \
998 score += name8(s, dst + 8, src + 8, stride, 8); \
1003 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)
1004 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)
1005 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)
1007 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)
1009 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)
1010 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
1011 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
1012 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
1014 int ff_check_alignment(void)
1016 static int did_fail = 0;
1017 LOCAL_ALIGNED_16(int, aligned, [4]);
1019 if ((intptr_t)aligned & 15) {
1021 #if HAVE_MMX || HAVE_ALTIVEC
1022 av_log(NULL, AV_LOG_ERROR,
1023 "Compiler did not align stack variables. Libavcodec has been miscompiled\n"
1024 "and may be very slow or crash. This is not a bug in libavcodec,\n"
1025 "but in the compiler. You may try recompiling using gcc >= 4.2.\n"
1026 "Do not report crashes to FFmpeg developers.\n");
1035 av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)
1037 ff_check_alignment();
1039 c->sum_abs_dctelem = sum_abs_dctelem_c;
1041 /* TODO [0] 16 [1] 8 */
1042 c->pix_abs[0][0] = pix_abs16_c;
1043 c->pix_abs[0][1] = pix_abs16_x2_c;
1044 c->pix_abs[0][2] = pix_abs16_y2_c;
1045 c->pix_abs[0][3] = pix_abs16_xy2_c;
1046 c->pix_abs[1][0] = pix_abs8_c;
1047 c->pix_abs[1][1] = pix_abs8_x2_c;
1048 c->pix_abs[1][2] = pix_abs8_y2_c;
1049 c->pix_abs[1][3] = pix_abs8_xy2_c;
1051 #define SET_CMP_FUNC(name) \
1052 c->name[0] = name ## 16_c; \
1053 c->name[1] = name ## 8x8_c;
1055 SET_CMP_FUNC(hadamard8_diff)
1056 c->hadamard8_diff[4] = hadamard8_intra16_c;
1057 c->hadamard8_diff[5] = hadamard8_intra8x8_c;
1058 SET_CMP_FUNC(dct_sad)
1059 SET_CMP_FUNC(dct_max)
1061 SET_CMP_FUNC(dct264_sad)
1063 c->sad[0] = pix_abs16_c;
1064 c->sad[1] = pix_abs8_c;
1065 c->sse[0] = sse16_c;
1068 SET_CMP_FUNC(quant_psnr)
1071 c->vsad[0] = vsad16_c;
1072 c->vsad[1] = vsad8_c;
1073 c->vsad[4] = vsad_intra16_c;
1074 c->vsad[5] = vsad_intra8_c;
1075 c->vsse[0] = vsse16_c;
1076 c->vsse[1] = vsse8_c;
1077 c->vsse[4] = vsse_intra16_c;
1078 c->vsse[5] = vsse_intra8_c;
1079 c->nsse[0] = nsse16_c;
1080 c->nsse[1] = nsse8_c;
1081 #if CONFIG_SNOW_DECODER || CONFIG_SNOW_ENCODER
1082 ff_dsputil_init_dwt(c);
1086 ff_me_cmp_init_alpha(c, avctx);
1088 ff_me_cmp_init_arm(c, avctx);
1090 ff_me_cmp_init_ppc(c, avctx);
1092 ff_me_cmp_init_x86(c, avctx);
1094 ff_me_cmp_init_mips(c, avctx);
1096 c->median_sad[0] = pix_median_abs16_c;
1097 c->median_sad[1] = pix_median_abs8_c;