cb9446d391e2874534b78fa9865e19ee15611d80
[libav.git] / libavcodec / ppc / fdct_altivec.c
1 /* ffmpeg/libavcodec/ppc/fdct_altivec.c, this file is part of the
2 * AltiVec optimized library for the FFMPEG Multimedia System
3 * Copyright (C) 2003 James Klicman <james@klicman.org>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22
23 #include "common.h"
24 #include "dsputil.h"
25 #include "dsputil_altivec.h"
26 #include "gcc_fixes.h"
27
28
29 #define vs16(v) ((vector signed short)(v))
30 #define vs32(v) ((vector signed int)(v))
31 #define vu8(v) ((vector unsigned char)(v))
32 #define vu16(v) ((vector unsigned short)(v))
33 #define vu32(v) ((vector unsigned int)(v))
34
35
36 #define C1 0.98078525066375732421875000 /* cos(1*PI/16) */
37 #define C2 0.92387950420379638671875000 /* cos(2*PI/16) */
38 #define C3 0.83146959543228149414062500 /* cos(3*PI/16) */
39 #define C4 0.70710676908493041992187500 /* cos(4*PI/16) */
40 #define C5 0.55557024478912353515625000 /* cos(5*PI/16) */
41 #define C6 0.38268342614173889160156250 /* cos(6*PI/16) */
42 #define C7 0.19509032368659973144531250 /* cos(7*PI/16) */
43 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */
44
45
46 #define W0 -(2 * C2)
47 #define W1 (2 * C6)
48 #define W2 (SQRT_2 * C6)
49 #define W3 (SQRT_2 * C3)
50 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))
51 #define W5 (SQRT_2 * ( C1 + C3 - C5 + C7))
52 #define W6 (SQRT_2 * ( C1 + C3 + C5 - C7))
53 #define W7 (SQRT_2 * ( C1 + C3 - C5 - C7))
54 #define W8 (SQRT_2 * ( C7 - C3))
55 #define W9 (SQRT_2 * (-C1 - C3))
56 #define WA (SQRT_2 * (-C3 - C5))
57 #define WB (SQRT_2 * ( C5 - C3))
58
59
60 static vector float fdctconsts[3] = {
61 (vector float)AVV( W0, W1, W2, W3 ),
62 (vector float)AVV( W4, W5, W6, W7 ),
63 (vector float)AVV( W8, W9, WA, WB )
64 };
65
66 #define LD_W0 vec_splat(cnsts0, 0)
67 #define LD_W1 vec_splat(cnsts0, 1)
68 #define LD_W2 vec_splat(cnsts0, 2)
69 #define LD_W3 vec_splat(cnsts0, 3)
70 #define LD_W4 vec_splat(cnsts1, 0)
71 #define LD_W5 vec_splat(cnsts1, 1)
72 #define LD_W6 vec_splat(cnsts1, 2)
73 #define LD_W7 vec_splat(cnsts1, 3)
74 #define LD_W8 vec_splat(cnsts2, 0)
75 #define LD_W9 vec_splat(cnsts2, 1)
76 #define LD_WA vec_splat(cnsts2, 2)
77 #define LD_WB vec_splat(cnsts2, 3)
78
79
80 #define FDCTROW(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \
81 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
82 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
83 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
84 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
85 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
86 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
87 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
88 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
89 \
90 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
91 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
92 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
93 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
94 \
95 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
96 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
97 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
98 cnst = LD_W2; \
99 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
100 cnst = LD_W1; \
101 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
102 cnst = LD_W0; \
103 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
104 \
105 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
106 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
107 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
108 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
109 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
110 cnst = LD_W3; \
111 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
112 \
113 cnst = LD_W8; \
114 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
115 cnst = LD_W9; \
116 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
117 cnst = LD_WA; \
118 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
119 cnst = LD_WB; \
120 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
121 \
122 cnst = LD_W4; \
123 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
124 cnst = LD_W5; \
125 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
126 cnst = LD_W6; \
127 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
128 cnst = LD_W7; \
129 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
130 \
131 b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \
132 b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \
133 b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \
134 b1 = vec_add(b1, x3); /* b1 = b1 + x3; */ \
135 /* }}} */
136
137 #define FDCTCOL(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \
138 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
139 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
140 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
141 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
142 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
143 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
144 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
145 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
146 \
147 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
148 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
149 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
150 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
151 \
152 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
153 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
154 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
155 cnst = LD_W2; \
156 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
157 cnst = LD_W1; \
158 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
159 cnst = LD_W0; \
160 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
161 \
162 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
163 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
164 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
165 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
166 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
167 cnst = LD_W3; \
168 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
169 \
170 cnst = LD_W8; \
171 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
172 cnst = LD_W9; \
173 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
174 cnst = LD_WA; \
175 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
176 cnst = LD_WB; \
177 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
178 \
179 cnst = LD_W4; \
180 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
181 cnst = LD_W5; \
182 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
183 cnst = LD_W6; \
184 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
185 cnst = LD_W7; \
186 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
187 \
188 b7 = vec_add(b7, x2); /* b7 += x2; */ \
189 b5 = vec_add(b5, x3); /* b5 += x3; */ \
190 b3 = vec_add(b3, x2); /* b3 += x2; */ \
191 b1 = vec_add(b1, x3); /* b1 += x3; */ \
192 /* }}} */
193
194
195
196 /* two dimensional discrete cosine transform */
197
198 void fdct_altivec(int16_t *block)
199 {
200 POWERPC_PERF_DECLARE(altivec_fdct, 1);
201 vector signed short *bp;
202 vector float *cp;
203 vector float b00, b10, b20, b30, b40, b50, b60, b70;
204 vector float b01, b11, b21, b31, b41, b51, b61, b71;
205 vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
206 vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
207
208 POWERPC_PERF_START_COUNT(altivec_fdct, 1);
209
210
211 /* setup constants {{{ */
212 /* mzero = -0.0 */
213 mzero = ((vector float)vec_splat_u32(-1));
214 mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero)));
215 cp = fdctconsts;
216 cnsts0 = vec_ld(0, cp); cp++;
217 cnsts1 = vec_ld(0, cp); cp++;
218 cnsts2 = vec_ld(0, cp);
219 /* }}} */
220
221
222 /* 8x8 matrix transpose (vector short[8]) {{{ */
223 #define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b))
224
225 bp = (vector signed short*)block;
226 b00 = ((vector float)vec_ld(0, bp));
227 b40 = ((vector float)vec_ld(16*4, bp));
228 b01 = ((vector float)MERGE_S16(h, b00, b40));
229 b11 = ((vector float)MERGE_S16(l, b00, b40));
230 bp++;
231 b10 = ((vector float)vec_ld(0, bp));
232 b50 = ((vector float)vec_ld(16*4, bp));
233 b21 = ((vector float)MERGE_S16(h, b10, b50));
234 b31 = ((vector float)MERGE_S16(l, b10, b50));
235 bp++;
236 b20 = ((vector float)vec_ld(0, bp));
237 b60 = ((vector float)vec_ld(16*4, bp));
238 b41 = ((vector float)MERGE_S16(h, b20, b60));
239 b51 = ((vector float)MERGE_S16(l, b20, b60));
240 bp++;
241 b30 = ((vector float)vec_ld(0, bp));
242 b70 = ((vector float)vec_ld(16*4, bp));
243 b61 = ((vector float)MERGE_S16(h, b30, b70));
244 b71 = ((vector float)MERGE_S16(l, b30, b70));
245
246 x0 = ((vector float)MERGE_S16(h, b01, b41));
247 x1 = ((vector float)MERGE_S16(l, b01, b41));
248 x2 = ((vector float)MERGE_S16(h, b11, b51));
249 x3 = ((vector float)MERGE_S16(l, b11, b51));
250 x4 = ((vector float)MERGE_S16(h, b21, b61));
251 x5 = ((vector float)MERGE_S16(l, b21, b61));
252 x6 = ((vector float)MERGE_S16(h, b31, b71));
253 x7 = ((vector float)MERGE_S16(l, b31, b71));
254
255 b00 = ((vector float)MERGE_S16(h, x0, x4));
256 b10 = ((vector float)MERGE_S16(l, x0, x4));
257 b20 = ((vector float)MERGE_S16(h, x1, x5));
258 b30 = ((vector float)MERGE_S16(l, x1, x5));
259 b40 = ((vector float)MERGE_S16(h, x2, x6));
260 b50 = ((vector float)MERGE_S16(l, x2, x6));
261 b60 = ((vector float)MERGE_S16(h, x3, x7));
262 b70 = ((vector float)MERGE_S16(l, x3, x7));
263
264 #undef MERGE_S16
265 /* }}} */
266
267
268 /* Some of the initial calculations can be done as vector short before
269 * conversion to vector float. The following code section takes advantage
270 * of this.
271 */
272 #if 1
273 /* fdct rows {{{ */
274 x0 = ((vector float)vec_add(vs16(b00), vs16(b70)));
275 x7 = ((vector float)vec_sub(vs16(b00), vs16(b70)));
276 x1 = ((vector float)vec_add(vs16(b10), vs16(b60)));
277 x6 = ((vector float)vec_sub(vs16(b10), vs16(b60)));
278 x2 = ((vector float)vec_add(vs16(b20), vs16(b50)));
279 x5 = ((vector float)vec_sub(vs16(b20), vs16(b50)));
280 x3 = ((vector float)vec_add(vs16(b30), vs16(b40)));
281 x4 = ((vector float)vec_sub(vs16(b30), vs16(b40)));
282
283 b70 = ((vector float)vec_add(vs16(x0), vs16(x3)));
284 b10 = ((vector float)vec_add(vs16(x1), vs16(x2)));
285
286 b00 = ((vector float)vec_add(vs16(b70), vs16(b10)));
287 b40 = ((vector float)vec_sub(vs16(b70), vs16(b10)));
288
289 #define CTF0(n) \
290 b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \
291 b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \
292 b##n##1 = vec_ctf(vs32(b##n##1), 0); \
293 b##n##0 = vec_ctf(vs32(b##n##0), 0);
294
295 CTF0(0);
296 CTF0(4);
297
298 b20 = ((vector float)vec_sub(vs16(x0), vs16(x3)));
299 b60 = ((vector float)vec_sub(vs16(x1), vs16(x2)));
300
301 CTF0(2);
302 CTF0(6);
303
304 #undef CTF0
305
306 x0 = vec_add(b60, b20);
307 x1 = vec_add(b61, b21);
308
309 cnst = LD_W2;
310 x0 = vec_madd(cnst, x0, mzero);
311 x1 = vec_madd(cnst, x1, mzero);
312 cnst = LD_W1;
313 b20 = vec_madd(cnst, b20, x0);
314 b21 = vec_madd(cnst, b21, x1);
315 cnst = LD_W0;
316 b60 = vec_madd(cnst, b60, x0);
317 b61 = vec_madd(cnst, b61, x1);
318
319 #define CTFX(x,b) \
320 b##0 = ((vector float)vec_unpackh(vs16(x))); \
321 b##1 = ((vector float)vec_unpackl(vs16(x))); \
322 b##0 = vec_ctf(vs32(b##0), 0); \
323 b##1 = vec_ctf(vs32(b##1), 0); \
324
325 CTFX(x4, b7);
326 CTFX(x5, b5);
327 CTFX(x6, b3);
328 CTFX(x7, b1);
329
330 #undef CTFX
331
332
333 x0 = vec_add(b70, b10);
334 x1 = vec_add(b50, b30);
335 x2 = vec_add(b70, b30);
336 x3 = vec_add(b50, b10);
337 x8 = vec_add(x2, x3);
338 cnst = LD_W3;
339 x8 = vec_madd(cnst, x8, mzero);
340
341 cnst = LD_W8;
342 x0 = vec_madd(cnst, x0, mzero);
343 cnst = LD_W9;
344 x1 = vec_madd(cnst, x1, mzero);
345 cnst = LD_WA;
346 x2 = vec_madd(cnst, x2, x8);
347 cnst = LD_WB;
348 x3 = vec_madd(cnst, x3, x8);
349
350 cnst = LD_W4;
351 b70 = vec_madd(cnst, b70, x0);
352 cnst = LD_W5;
353 b50 = vec_madd(cnst, b50, x1);
354 cnst = LD_W6;
355 b30 = vec_madd(cnst, b30, x1);
356 cnst = LD_W7;
357 b10 = vec_madd(cnst, b10, x0);
358
359 b70 = vec_add(b70, x2);
360 b50 = vec_add(b50, x3);
361 b30 = vec_add(b30, x2);
362 b10 = vec_add(b10, x3);
363
364
365 x0 = vec_add(b71, b11);
366 x1 = vec_add(b51, b31);
367 x2 = vec_add(b71, b31);
368 x3 = vec_add(b51, b11);
369 x8 = vec_add(x2, x3);
370 cnst = LD_W3;
371 x8 = vec_madd(cnst, x8, mzero);
372
373 cnst = LD_W8;
374 x0 = vec_madd(cnst, x0, mzero);
375 cnst = LD_W9;
376 x1 = vec_madd(cnst, x1, mzero);
377 cnst = LD_WA;
378 x2 = vec_madd(cnst, x2, x8);
379 cnst = LD_WB;
380 x3 = vec_madd(cnst, x3, x8);
381
382 cnst = LD_W4;
383 b71 = vec_madd(cnst, b71, x0);
384 cnst = LD_W5;
385 b51 = vec_madd(cnst, b51, x1);
386 cnst = LD_W6;
387 b31 = vec_madd(cnst, b31, x1);
388 cnst = LD_W7;
389 b11 = vec_madd(cnst, b11, x0);
390
391 b71 = vec_add(b71, x2);
392 b51 = vec_add(b51, x3);
393 b31 = vec_add(b31, x2);
394 b11 = vec_add(b11, x3);
395 /* }}} */
396 #else
397 /* convert to float {{{ */
398 #define CTF(n) \
399 vs32(b##n##1) = vec_unpackl(vs16(b##n##0)); \
400 vs32(b##n##0) = vec_unpackh(vs16(b##n##0)); \
401 b##n##1 = vec_ctf(vs32(b##n##1), 0); \
402 b##n##0 = vec_ctf(vs32(b##n##0), 0); \
403
404 CTF(0);
405 CTF(1);
406 CTF(2);
407 CTF(3);
408 CTF(4);
409 CTF(5);
410 CTF(6);
411 CTF(7);
412
413 #undef CTF
414 /* }}} */
415
416 FDCTROW(b00, b10, b20, b30, b40, b50, b60, b70);
417 FDCTROW(b01, b11, b21, b31, b41, b51, b61, b71);
418 #endif
419
420
421 /* 8x8 matrix transpose (vector float[8][2]) {{{ */
422 x0 = vec_mergel(b00, b20);
423 x1 = vec_mergeh(b00, b20);
424 x2 = vec_mergel(b10, b30);
425 x3 = vec_mergeh(b10, b30);
426
427 b00 = vec_mergeh(x1, x3);
428 b10 = vec_mergel(x1, x3);
429 b20 = vec_mergeh(x0, x2);
430 b30 = vec_mergel(x0, x2);
431
432 x4 = vec_mergel(b41, b61);
433 x5 = vec_mergeh(b41, b61);
434 x6 = vec_mergel(b51, b71);
435 x7 = vec_mergeh(b51, b71);
436
437 b41 = vec_mergeh(x5, x7);
438 b51 = vec_mergel(x5, x7);
439 b61 = vec_mergeh(x4, x6);
440 b71 = vec_mergel(x4, x6);
441
442 x0 = vec_mergel(b01, b21);
443 x1 = vec_mergeh(b01, b21);
444 x2 = vec_mergel(b11, b31);
445 x3 = vec_mergeh(b11, b31);
446
447 x4 = vec_mergel(b40, b60);
448 x5 = vec_mergeh(b40, b60);
449 x6 = vec_mergel(b50, b70);
450 x7 = vec_mergeh(b50, b70);
451
452 b40 = vec_mergeh(x1, x3);
453 b50 = vec_mergel(x1, x3);
454 b60 = vec_mergeh(x0, x2);
455 b70 = vec_mergel(x0, x2);
456
457 b01 = vec_mergeh(x5, x7);
458 b11 = vec_mergel(x5, x7);
459 b21 = vec_mergeh(x4, x6);
460 b31 = vec_mergel(x4, x6);
461 /* }}} */
462
463
464 FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
465 FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
466
467
468 /* round, convert back to short {{{ */
469 #define CTS(n) \
470 b##n##0 = vec_round(b##n##0); \
471 b##n##1 = vec_round(b##n##1); \
472 b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \
473 b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \
474 b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \
475 vec_st(vs16(b##n##0), 0, bp);
476
477 bp = (vector signed short*)block;
478 CTS(0); bp++;
479 CTS(1); bp++;
480 CTS(2); bp++;
481 CTS(3); bp++;
482 CTS(4); bp++;
483 CTS(5); bp++;
484 CTS(6); bp++;
485 CTS(7);
486
487 #undef CTS
488 /* }}} */
489
490 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
491 }
492
493 /* vim:set foldmethod=marker foldlevel=0: */