FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
fdct_altivec.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2003 James Klicman <james@klicman.org>
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 #include "config.h"
22 #if HAVE_ALTIVEC_H
23 #include <altivec.h>
24 #endif
25 
26 #include "libavutil/common.h"
27 #include "dsputil_altivec.h"
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 #define C1 0.98078525066375732421875000 /* cos(1 * PI / 16) */
36 #define C2 0.92387950420379638671875000 /* cos(2 * PI / 16) */
37 #define C3 0.83146959543228149414062500 /* cos(3 * PI / 16) */
38 #define C4 0.70710676908493041992187500 /* cos(4 * PI / 16) */
39 #define C5 0.55557024478912353515625000 /* cos(5 * PI / 16) */
40 #define C6 0.38268342614173889160156250 /* cos(6 * PI / 16) */
41 #define C7 0.19509032368659973144531250 /* cos(7 * PI / 16) */
42 #define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */
43 
44 #define W0 -(2 * C2)
45 #define W1 (2 * C6)
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))
56 
57 static vector float fdctconsts[3] = {
58  { W0, W1, W2, W3 },
59  { W4, W5, W6, W7 },
60  { W8, W9, WA, WB }
61 };
62 
63 #define LD_W0 vec_splat(cnsts0, 0)
64 #define LD_W1 vec_splat(cnsts0, 1)
65 #define LD_W2 vec_splat(cnsts0, 2)
66 #define LD_W3 vec_splat(cnsts0, 3)
67 #define LD_W4 vec_splat(cnsts1, 0)
68 #define LD_W5 vec_splat(cnsts1, 1)
69 #define LD_W6 vec_splat(cnsts1, 2)
70 #define LD_W7 vec_splat(cnsts1, 3)
71 #define LD_W8 vec_splat(cnsts2, 0)
72 #define LD_W9 vec_splat(cnsts2, 1)
73 #define LD_WA vec_splat(cnsts2, 2)
74 #define LD_WB vec_splat(cnsts2, 3)
75 
76 #define FDCTROW(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \
77  x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
78  x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
79  x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
80  x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
81  x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
82  x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
83  x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
84  x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
85  \
86  b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
87  b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
88  b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
89  b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
90  \
91  b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
92  b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
93  b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
94  cnst = LD_W2; \
95  b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
96  cnst = LD_W1; \
97  b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
98  cnst = LD_W0; \
99  b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
100  \
101  x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
102  x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
103  x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
104  x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
105  x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
106  cnst = LD_W3; \
107  x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
108  \
109  cnst = LD_W8; \
110  x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
111  cnst = LD_W9; \
112  x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
113  cnst = LD_WA; \
114  x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
115  cnst = LD_WB; \
116  x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
117  \
118  cnst = LD_W4; \
119  b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
120  cnst = LD_W5; \
121  b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
122  cnst = LD_W6; \
123  b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
124  cnst = LD_W7; \
125  b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
126  \
127  b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \
128  b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \
129  b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \
130  b1 = vec_add(b1, x3) /* b1 = b1 + x3; */ \
131  /* }}} */
132 
133 #define FDCTCOL(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \
134  x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \
135  x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \
136  x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \
137  x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \
138  x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \
139  x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \
140  x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \
141  x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \
142  \
143  b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \
144  b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \
145  b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \
146  b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \
147  \
148  b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \
149  b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \
150  b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \
151  cnst = LD_W2; \
152  b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \
153  cnst = LD_W1; \
154  b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \
155  cnst = LD_W0; \
156  b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \
157  \
158  x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \
159  x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \
160  x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \
161  x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \
162  x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \
163  cnst = LD_W3; \
164  x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \
165  \
166  cnst = LD_W8; \
167  x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \
168  cnst = LD_W9; \
169  x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \
170  cnst = LD_WA; \
171  x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \
172  cnst = LD_WB; \
173  x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \
174  \
175  cnst = LD_W4; \
176  b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \
177  cnst = LD_W5; \
178  b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \
179  cnst = LD_W6; \
180  b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \
181  cnst = LD_W7; \
182  b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \
183  \
184  b7 = vec_add(b7, x2); /* b7 += x2; */ \
185  b5 = vec_add(b5, x3); /* b5 += x3; */ \
186  b3 = vec_add(b3, x2); /* b3 += x2; */ \
187  b1 = vec_add(b1, x3) /* b1 += x3; */ \
188  /* }}} */
189 
190 /* two dimensional discrete cosine transform */
191 void ff_fdct_altivec(int16_t *block)
192 {
193  vector signed short *bp;
194  vector float *cp = fdctconsts;
195  vector float b00, b10, b20, b30, b40, b50, b60, b70;
196  vector float b01, b11, b21, b31, b41, b51, b61, b71;
197  vector float mzero, cnst, cnsts0, cnsts1, cnsts2;
198  vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;
199 
200  /* setup constants {{{ */
201  /* mzero = -0.0 */
202  mzero = ((vector float) vec_splat_u32(-1));
203  mzero = ((vector float) vec_sl(vu32(mzero), vu32(mzero)));
204  cnsts0 = vec_ld(0, cp);
205  cp++;
206  cnsts1 = vec_ld(0, cp);
207  cp++;
208  cnsts2 = vec_ld(0, cp);
209  /* }}} */
210 
211  /* 8x8 matrix transpose (vector short[8]) {{{ */
212 #define MERGE_S16(hl, a, b) vec_merge ## hl(vs16(a), vs16(b))
213 
214  bp = (vector signed short *) block;
215  b00 = ((vector float) vec_ld(0, bp));
216  b40 = ((vector float) vec_ld(16 * 4, bp));
217  b01 = ((vector float) MERGE_S16(h, b00, b40));
218  b11 = ((vector float) MERGE_S16(l, b00, b40));
219  bp++;
220  b10 = ((vector float) vec_ld(0, bp));
221  b50 = ((vector float) vec_ld(16 * 4, bp));
222  b21 = ((vector float) MERGE_S16(h, b10, b50));
223  b31 = ((vector float) MERGE_S16(l, b10, b50));
224  bp++;
225  b20 = ((vector float) vec_ld(0, bp));
226  b60 = ((vector float) vec_ld(16 * 4, bp));
227  b41 = ((vector float) MERGE_S16(h, b20, b60));
228  b51 = ((vector float) MERGE_S16(l, b20, b60));
229  bp++;
230  b30 = ((vector float) vec_ld(0, bp));
231  b70 = ((vector float) vec_ld(16 * 4, bp));
232  b61 = ((vector float) MERGE_S16(h, b30, b70));
233  b71 = ((vector float) MERGE_S16(l, b30, b70));
234 
235  x0 = ((vector float) MERGE_S16(h, b01, b41));
236  x1 = ((vector float) MERGE_S16(l, b01, b41));
237  x2 = ((vector float) MERGE_S16(h, b11, b51));
238  x3 = ((vector float) MERGE_S16(l, b11, b51));
239  x4 = ((vector float) MERGE_S16(h, b21, b61));
240  x5 = ((vector float) MERGE_S16(l, b21, b61));
241  x6 = ((vector float) MERGE_S16(h, b31, b71));
242  x7 = ((vector float) MERGE_S16(l, b31, b71));
243 
244  b00 = ((vector float) MERGE_S16(h, x0, x4));
245  b10 = ((vector float) MERGE_S16(l, x0, x4));
246  b20 = ((vector float) MERGE_S16(h, x1, x5));
247  b30 = ((vector float) MERGE_S16(l, x1, x5));
248  b40 = ((vector float) MERGE_S16(h, x2, x6));
249  b50 = ((vector float) MERGE_S16(l, x2, x6));
250  b60 = ((vector float) MERGE_S16(h, x3, x7));
251  b70 = ((vector float) MERGE_S16(l, x3, x7));
252 
253 #undef MERGE_S16
254  /* }}} */
255 
256  /* Some of the initial calculations can be done as vector short
257  * before conversion to vector float. The following code section
258  * takes advantage of this. */
259 
260  /* fdct rows {{{ */
261  x0 = ((vector float) vec_add(vs16(b00), vs16(b70)));
262  x7 = ((vector float) vec_sub(vs16(b00), vs16(b70)));
263  x1 = ((vector float) vec_add(vs16(b10), vs16(b60)));
264  x6 = ((vector float) vec_sub(vs16(b10), vs16(b60)));
265  x2 = ((vector float) vec_add(vs16(b20), vs16(b50)));
266  x5 = ((vector float) vec_sub(vs16(b20), vs16(b50)));
267  x3 = ((vector float) vec_add(vs16(b30), vs16(b40)));
268  x4 = ((vector float) vec_sub(vs16(b30), vs16(b40)));
269 
270  b70 = ((vector float) vec_add(vs16(x0), vs16(x3)));
271  b10 = ((vector float) vec_add(vs16(x1), vs16(x2)));
272 
273  b00 = ((vector float) vec_add(vs16(b70), vs16(b10)));
274  b40 = ((vector float) vec_sub(vs16(b70), vs16(b10)));
275 
276 #define CTF0(n) \
277  b ## n ## 1 = ((vector float) vec_unpackl(vs16(b ## n ## 0))); \
278  b ## n ## 0 = ((vector float) vec_unpackh(vs16(b ## n ## 0))); \
279  b ## n ## 1 = vec_ctf(vs32(b ## n ## 1), 0); \
280  b ## n ## 0 = vec_ctf(vs32(b ## n ## 0), 0)
281 
282  CTF0(0);
283  CTF0(4);
284 
285  b20 = ((vector float) vec_sub(vs16(x0), vs16(x3)));
286  b60 = ((vector float) vec_sub(vs16(x1), vs16(x2)));
287 
288  CTF0(2);
289  CTF0(6);
290 
291 #undef CTF0
292 
293  x0 = vec_add(b60, b20);
294  x1 = vec_add(b61, b21);
295 
296  cnst = LD_W2;
297  x0 = vec_madd(cnst, x0, mzero);
298  x1 = vec_madd(cnst, x1, mzero);
299  cnst = LD_W1;
300  b20 = vec_madd(cnst, b20, x0);
301  b21 = vec_madd(cnst, b21, x1);
302  cnst = LD_W0;
303  b60 = vec_madd(cnst, b60, x0);
304  b61 = vec_madd(cnst, b61, x1);
305 
306 #define CTFX(x, b) \
307  b ## 0 = ((vector float) vec_unpackh(vs16(x))); \
308  b ## 1 = ((vector float) vec_unpackl(vs16(x))); \
309  b ## 0 = vec_ctf(vs32(b ## 0), 0); \
310  b ## 1 = vec_ctf(vs32(b ## 1), 0)
311 
312  CTFX(x4, b7);
313  CTFX(x5, b5);
314  CTFX(x6, b3);
315  CTFX(x7, b1);
316 
317 #undef CTFX
318 
319  x0 = vec_add(b70, b10);
320  x1 = vec_add(b50, b30);
321  x2 = vec_add(b70, b30);
322  x3 = vec_add(b50, b10);
323  x8 = vec_add(x2, x3);
324  cnst = LD_W3;
325  x8 = vec_madd(cnst, x8, mzero);
326 
327  cnst = LD_W8;
328  x0 = vec_madd(cnst, x0, mzero);
329  cnst = LD_W9;
330  x1 = vec_madd(cnst, x1, mzero);
331  cnst = LD_WA;
332  x2 = vec_madd(cnst, x2, x8);
333  cnst = LD_WB;
334  x3 = vec_madd(cnst, x3, x8);
335 
336  cnst = LD_W4;
337  b70 = vec_madd(cnst, b70, x0);
338  cnst = LD_W5;
339  b50 = vec_madd(cnst, b50, x1);
340  cnst = LD_W6;
341  b30 = vec_madd(cnst, b30, x1);
342  cnst = LD_W7;
343  b10 = vec_madd(cnst, b10, x0);
344 
345  b70 = vec_add(b70, x2);
346  b50 = vec_add(b50, x3);
347  b30 = vec_add(b30, x2);
348  b10 = vec_add(b10, x3);
349 
350  x0 = vec_add(b71, b11);
351  x1 = vec_add(b51, b31);
352  x2 = vec_add(b71, b31);
353  x3 = vec_add(b51, b11);
354  x8 = vec_add(x2, x3);
355  cnst = LD_W3;
356  x8 = vec_madd(cnst, x8, mzero);
357 
358  cnst = LD_W8;
359  x0 = vec_madd(cnst, x0, mzero);
360  cnst = LD_W9;
361  x1 = vec_madd(cnst, x1, mzero);
362  cnst = LD_WA;
363  x2 = vec_madd(cnst, x2, x8);
364  cnst = LD_WB;
365  x3 = vec_madd(cnst, x3, x8);
366 
367  cnst = LD_W4;
368  b71 = vec_madd(cnst, b71, x0);
369  cnst = LD_W5;
370  b51 = vec_madd(cnst, b51, x1);
371  cnst = LD_W6;
372  b31 = vec_madd(cnst, b31, x1);
373  cnst = LD_W7;
374  b11 = vec_madd(cnst, b11, x0);
375 
376  b71 = vec_add(b71, x2);
377  b51 = vec_add(b51, x3);
378  b31 = vec_add(b31, x2);
379  b11 = vec_add(b11, x3);
380  /* }}} */
381 
382  /* 8x8 matrix transpose (vector float[8][2]) {{{ */
383  x0 = vec_mergel(b00, b20);
384  x1 = vec_mergeh(b00, b20);
385  x2 = vec_mergel(b10, b30);
386  x3 = vec_mergeh(b10, b30);
387 
388  b00 = vec_mergeh(x1, x3);
389  b10 = vec_mergel(x1, x3);
390  b20 = vec_mergeh(x0, x2);
391  b30 = vec_mergel(x0, x2);
392 
393  x4 = vec_mergel(b41, b61);
394  x5 = vec_mergeh(b41, b61);
395  x6 = vec_mergel(b51, b71);
396  x7 = vec_mergeh(b51, b71);
397 
398  b41 = vec_mergeh(x5, x7);
399  b51 = vec_mergel(x5, x7);
400  b61 = vec_mergeh(x4, x6);
401  b71 = vec_mergel(x4, x6);
402 
403  x0 = vec_mergel(b01, b21);
404  x1 = vec_mergeh(b01, b21);
405  x2 = vec_mergel(b11, b31);
406  x3 = vec_mergeh(b11, b31);
407 
408  x4 = vec_mergel(b40, b60);
409  x5 = vec_mergeh(b40, b60);
410  x6 = vec_mergel(b50, b70);
411  x7 = vec_mergeh(b50, b70);
412 
413  b40 = vec_mergeh(x1, x3);
414  b50 = vec_mergel(x1, x3);
415  b60 = vec_mergeh(x0, x2);
416  b70 = vec_mergel(x0, x2);
417 
418  b01 = vec_mergeh(x5, x7);
419  b11 = vec_mergel(x5, x7);
420  b21 = vec_mergeh(x4, x6);
421  b31 = vec_mergel(x4, x6);
422  /* }}} */
423 
424  FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
425  FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
426 
427  /* round, convert back to short {{{ */
428 #define CTS(n) \
429  b ## n ## 0 = vec_round(b ## n ## 0); \
430  b ## n ## 1 = vec_round(b ## n ## 1); \
431  b ## n ## 0 = ((vector float) vec_cts(b ## n ## 0, 0)); \
432  b ## n ## 1 = ((vector float) vec_cts(b ## n ## 1, 0)); \
433  b ## n ## 0 = ((vector float) vec_pack(vs32(b ## n ## 0), \
434  vs32(b ## n ## 1))); \
435  vec_st(vs16(b ## n ## 0), 0, bp)
436 
437  bp = (vector signed short *) block;
438  CTS(0);
439  bp++;
440  CTS(1);
441  bp++;
442  CTS(2);
443  bp++;
444  CTS(3);
445  bp++;
446  CTS(4);
447  bp++;
448  CTS(5);
449  bp++;
450  CTS(6);
451  bp++;
452  CTS(7);
453 
454 #undef CTS
455  /* }}} */
456 }
Generated on Mon Jul 7 2014 19:21:08 for FFmpeg by   doxygen 1.8.2