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>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2.1 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "../dsputil.h"
23 #include "dsputil_altivec.h"
24 #include "gcc_fixes.h"
27 #define vs16(v) ((vector signed short)(v))
28 #define vs32(v) ((vector signed int)(v))
29 #define vu8(v) ((vector unsigned char)(v))
30 #define vu16(v) ((vector unsigned short)(v))
31 #define vu32(v) ((vector unsigned int)(v))
34 #define C1 0.98078525066375732421875000 /* cos(1*PI/16) */
35 #define C2 0.92387950420379638671875000 /* cos(2*PI/16) */
36 #define C3 0.83146959543228149414062500 /* cos(3*PI/16) */
37 #define C4 0.70710676908493041992187500 /* cos(4*PI/16) */
38 #define C5 0.55557024478912353515625000 /* cos(5*PI/16) */
39 #define C6 0.38268342614173889160156250 /* cos(6*PI/16) */
40 #define C7 0.19509032368659973144531250 /* cos(7*PI/16) */
41 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */
46 #define W2 (SQRT_2 * C6)
47 #define W3 (SQRT_2 * C3)
48 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))
49 #define W5 (SQRT_2 * ( C1 + C3 - C5 + C7))
50 #define W6 (SQRT_2 * ( C1 + C3 + C5 - C7))
51 #define W7 (SQRT_2 * ( C1 + C3 - C5 - C7))
52 #define W8 (SQRT_2 * ( C7 - C3))
53 #define W9 (SQRT_2 * (-C1 - C3))
54 #define WA (SQRT_2 * (-C3 - C5))
55 #define WB (SQRT_2 * ( C5 - C3))
58 static vector float fdctconsts[3] = {
59 (vector float)AVV( W0, W1, W2, W3 ),
60 (vector float)AVV( W4, W5, W6, W7 ),
61 (vector float)AVV( W8, W9, WA, WB )
64 #define LD_W0 vec_splat(cnsts0, 0)
65 #define LD_W1 vec_splat(cnsts0, 1)
66 #define LD_W2 vec_splat(cnsts0, 2)
67 #define LD_W3 vec_splat(cnsts0, 3)
68 #define LD_W4 vec_splat(cnsts1, 0)
69 #define LD_W5 vec_splat(cnsts1, 1)
70 #define LD_W6 vec_splat(cnsts1, 2)
71 #define LD_W7 vec_splat(cnsts1, 3)
72 #define LD_W8 vec_splat(cnsts2, 0)
73 #define LD_W9 vec_splat(cnsts2, 1)
74 #define LD_WA vec_splat(cnsts2, 2)
75 #define LD_WB vec_splat(cnsts2, 3)
78 #define FDCTROW(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \
79 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
80 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
81 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
82 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
83 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
84 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
85 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
86 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
88 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
89 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
90 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
91 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
93 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
94 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
95 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
97 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
99 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
101 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
103 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
104 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
105 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
106 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
107 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
109 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
112 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
114 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
116 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
118 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
121 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
123 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
125 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
127 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
129 b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \
130 b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \
131 b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \
132 b1 = vec_add(b1, x3); /* b1 = b1 + x3; */ \
135 #define FDCTCOL(b0,b1,b2,b3,b4,b5,b6,b7) /* {{{ */ \
136 x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
137 x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
138 x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
139 x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
140 x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
141 x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
142 x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
143 x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
145 b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
146 b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
147 b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
148 b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
150 b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
151 b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
152 b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
154 b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
156 b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
158 b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
160 x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
161 x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
162 x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
163 x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
164 x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
166 x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
169 x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
171 x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
173 x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
175 x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
178 b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
180 b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
182 b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
184 b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
186 b7 = vec_add(b7, x2); /* b7 += x2; */ \
187 b5 = vec_add(b5, x3); /* b5 += x3; */ \
188 b3 = vec_add(b3, x2); /* b3 += x2; */ \
189 b1 = vec_add(b1, x3); /* b1 += x3; */ \
194 /* two dimensional discrete cosine transform */
196 void fdct_altivec(int16_t *block)
198 POWERPC_PERF_DECLARE(altivec_fdct, 1);
199 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
200 POWERPC_PERF_START_COUNT(altivec_fdct, 1);
201 void ff_jpeg_fdct_islow(int16_t *block);
202 ff_jpeg_fdct_islow(block);
203 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
204 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
205 vector signed short *bp;
207 vector float b00, b10, b20, b30, b40, b50, b60, b70;
208 vector float b01, b11, b21, b31, b41, b51, b61, b71;
209 vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
210 vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
212 POWERPC_PERF_START_COUNT(altivec_fdct, 1);
215 /* setup constants {{{ */
217 mzero = ((vector float)vec_splat_u32(-1));
218 mzero = ((vector float)vec_sl(vu32(mzero), vu32(mzero)));
220 cnsts0 = vec_ld(0, cp); cp++;
221 cnsts1 = vec_ld(0, cp); cp++;
222 cnsts2 = vec_ld(0, cp);
226 /* 8x8 matrix transpose (vector short[8]) {{{ */
227 #define MERGE_S16(hl,a,b) vec_merge##hl(vs16(a), vs16(b))
229 bp = (vector signed short*)block;
230 b00 = ((vector float)vec_ld(0, bp));
231 b40 = ((vector float)vec_ld(16*4, bp));
232 b01 = ((vector float)MERGE_S16(h, b00, b40));
233 b11 = ((vector float)MERGE_S16(l, b00, b40));
235 b10 = ((vector float)vec_ld(0, bp));
236 b50 = ((vector float)vec_ld(16*4, bp));
237 b21 = ((vector float)MERGE_S16(h, b10, b50));
238 b31 = ((vector float)MERGE_S16(l, b10, b50));
240 b20 = ((vector float)vec_ld(0, bp));
241 b60 = ((vector float)vec_ld(16*4, bp));
242 b41 = ((vector float)MERGE_S16(h, b20, b60));
243 b51 = ((vector float)MERGE_S16(l, b20, b60));
245 b30 = ((vector float)vec_ld(0, bp));
246 b70 = ((vector float)vec_ld(16*4, bp));
247 b61 = ((vector float)MERGE_S16(h, b30, b70));
248 b71 = ((vector float)MERGE_S16(l, b30, b70));
250 x0 = ((vector float)MERGE_S16(h, b01, b41));
251 x1 = ((vector float)MERGE_S16(l, b01, b41));
252 x2 = ((vector float)MERGE_S16(h, b11, b51));
253 x3 = ((vector float)MERGE_S16(l, b11, b51));
254 x4 = ((vector float)MERGE_S16(h, b21, b61));
255 x5 = ((vector float)MERGE_S16(l, b21, b61));
256 x6 = ((vector float)MERGE_S16(h, b31, b71));
257 x7 = ((vector float)MERGE_S16(l, b31, b71));
259 b00 = ((vector float)MERGE_S16(h, x0, x4));
260 b10 = ((vector float)MERGE_S16(l, x0, x4));
261 b20 = ((vector float)MERGE_S16(h, x1, x5));
262 b30 = ((vector float)MERGE_S16(l, x1, x5));
263 b40 = ((vector float)MERGE_S16(h, x2, x6));
264 b50 = ((vector float)MERGE_S16(l, x2, x6));
265 b60 = ((vector float)MERGE_S16(h, x3, x7));
266 b70 = ((vector float)MERGE_S16(l, x3, x7));
272 /* Some of the initial calculations can be done as vector short before
273 * conversion to vector float. The following code section takes advantage
278 x0 = ((vector float)vec_add(vs16(b00), vs16(b70)));
279 x7 = ((vector float)vec_sub(vs16(b00), vs16(b70)));
280 x1 = ((vector float)vec_add(vs16(b10), vs16(b60)));
281 x6 = ((vector float)vec_sub(vs16(b10), vs16(b60)));
282 x2 = ((vector float)vec_add(vs16(b20), vs16(b50)));
283 x5 = ((vector float)vec_sub(vs16(b20), vs16(b50)));
284 x3 = ((vector float)vec_add(vs16(b30), vs16(b40)));
285 x4 = ((vector float)vec_sub(vs16(b30), vs16(b40)));
287 b70 = ((vector float)vec_add(vs16(x0), vs16(x3)));
288 b10 = ((vector float)vec_add(vs16(x1), vs16(x2)));
290 b00 = ((vector float)vec_add(vs16(b70), vs16(b10)));
291 b40 = ((vector float)vec_sub(vs16(b70), vs16(b10)));
294 b##n##1 = ((vector float)vec_unpackl(vs16(b##n##0))); \
295 b##n##0 = ((vector float)vec_unpackh(vs16(b##n##0))); \
296 b##n##1 = vec_ctf(vs32(b##n##1), 0); \
297 b##n##0 = vec_ctf(vs32(b##n##0), 0);
302 b20 = ((vector float)vec_sub(vs16(x0), vs16(x3)));
303 b60 = ((vector float)vec_sub(vs16(x1), vs16(x2)));
310 x0 = vec_add(b60, b20);
311 x1 = vec_add(b61, b21);
314 x0 = vec_madd(cnst, x0, mzero);
315 x1 = vec_madd(cnst, x1, mzero);
317 b20 = vec_madd(cnst, b20, x0);
318 b21 = vec_madd(cnst, b21, x1);
320 b60 = vec_madd(cnst, b60, x0);
321 b61 = vec_madd(cnst, b61, x1);
324 b##0 = ((vector float)vec_unpackh(vs16(x))); \
325 b##1 = ((vector float)vec_unpackl(vs16(x))); \
326 b##0 = vec_ctf(vs32(b##0), 0); \
327 b##1 = vec_ctf(vs32(b##1), 0); \
337 x0 = vec_add(b70, b10);
338 x1 = vec_add(b50, b30);
339 x2 = vec_add(b70, b30);
340 x3 = vec_add(b50, b10);
341 x8 = vec_add(x2, x3);
343 x8 = vec_madd(cnst, x8, mzero);
346 x0 = vec_madd(cnst, x0, mzero);
348 x1 = vec_madd(cnst, x1, mzero);
350 x2 = vec_madd(cnst, x2, x8);
352 x3 = vec_madd(cnst, x3, x8);
355 b70 = vec_madd(cnst, b70, x0);
357 b50 = vec_madd(cnst, b50, x1);
359 b30 = vec_madd(cnst, b30, x1);
361 b10 = vec_madd(cnst, b10, x0);
363 b70 = vec_add(b70, x2);
364 b50 = vec_add(b50, x3);
365 b30 = vec_add(b30, x2);
366 b10 = vec_add(b10, x3);
369 x0 = vec_add(b71, b11);
370 x1 = vec_add(b51, b31);
371 x2 = vec_add(b71, b31);
372 x3 = vec_add(b51, b11);
373 x8 = vec_add(x2, x3);
375 x8 = vec_madd(cnst, x8, mzero);
378 x0 = vec_madd(cnst, x0, mzero);
380 x1 = vec_madd(cnst, x1, mzero);
382 x2 = vec_madd(cnst, x2, x8);
384 x3 = vec_madd(cnst, x3, x8);
387 b71 = vec_madd(cnst, b71, x0);
389 b51 = vec_madd(cnst, b51, x1);
391 b31 = vec_madd(cnst, b31, x1);
393 b11 = vec_madd(cnst, b11, x0);
395 b71 = vec_add(b71, x2);
396 b51 = vec_add(b51, x3);
397 b31 = vec_add(b31, x2);
398 b11 = vec_add(b11, x3);
401 /* convert to float {{{ */
403 vs32(b##n##1) = vec_unpackl(vs16(b##n##0)); \
404 vs32(b##n##0) = vec_unpackh(vs16(b##n##0)); \
405 b##n##1 = vec_ctf(vs32(b##n##1), 0); \
406 b##n##0 = vec_ctf(vs32(b##n##0), 0); \
420 FDCTROW(b00, b10, b20, b30, b40, b50, b60, b70);
421 FDCTROW(b01, b11, b21, b31, b41, b51, b61, b71);
425 /* 8x8 matrix transpose (vector float[8][2]) {{{ */
426 x0 = vec_mergel(b00, b20);
427 x1 = vec_mergeh(b00, b20);
428 x2 = vec_mergel(b10, b30);
429 x3 = vec_mergeh(b10, b30);
431 b00 = vec_mergeh(x1, x3);
432 b10 = vec_mergel(x1, x3);
433 b20 = vec_mergeh(x0, x2);
434 b30 = vec_mergel(x0, x2);
436 x4 = vec_mergel(b41, b61);
437 x5 = vec_mergeh(b41, b61);
438 x6 = vec_mergel(b51, b71);
439 x7 = vec_mergeh(b51, b71);
441 b41 = vec_mergeh(x5, x7);
442 b51 = vec_mergel(x5, x7);
443 b61 = vec_mergeh(x4, x6);
444 b71 = vec_mergel(x4, x6);
446 x0 = vec_mergel(b01, b21);
447 x1 = vec_mergeh(b01, b21);
448 x2 = vec_mergel(b11, b31);
449 x3 = vec_mergeh(b11, b31);
451 x4 = vec_mergel(b40, b60);
452 x5 = vec_mergeh(b40, b60);
453 x6 = vec_mergel(b50, b70);
454 x7 = vec_mergeh(b50, b70);
456 b40 = vec_mergeh(x1, x3);
457 b50 = vec_mergel(x1, x3);
458 b60 = vec_mergeh(x0, x2);
459 b70 = vec_mergel(x0, x2);
461 b01 = vec_mergeh(x5, x7);
462 b11 = vec_mergel(x5, x7);
463 b21 = vec_mergeh(x4, x6);
464 b31 = vec_mergel(x4, x6);
468 FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
469 FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
472 /* round, convert back to short {{{ */
474 b##n##0 = vec_round(b##n##0); \
475 b##n##1 = vec_round(b##n##1); \
476 b##n##0 = ((vector float)vec_cts(b##n##0, 0)); \
477 b##n##1 = ((vector float)vec_cts(b##n##1, 0)); \
478 b##n##0 = ((vector float)vec_pack(vs32(b##n##0), vs32(b##n##1))); \
479 vec_st(vs16(b##n##0), 0, bp);
481 bp = (vector signed short*)block;
494 POWERPC_PERF_STOP_COUNT(altivec_fdct, 1);
495 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
498 /* vim:set foldmethod=marker foldlevel=0: */