1 --- include/a52.h 2006-06-12 15:04:57.000000000 +0200
2 +++ liba52/a52.h 2006-06-05 02:23:02.000000000 +0200
4 int a52_block (a52_state_t * state);
5 void a52_free (a52_state_t * state);
7 +void* a52_resample_init(uint32_t mm_accel,int flags,int chans);
8 +extern int (* a52_resample) (float * _f, int16_t * s16);
10 +uint16_t crc16_block(uint8_t *data,uint32_t num_bytes);
13 --- liba52/a52_internal.h 2006-06-12 15:05:07.000000000 +0200
14 +++ liba52/a52_internal.h 2006-06-05 02:23:02.000000000 +0200
16 #define DELTA_BIT_NONE (2)
17 #define DELTA_BIT_RESERVED (3)
24 +# define REG_BP "rbp"
30 +# define REG_BP "ebp"
33 void a52_bit_allocate (a52_state_t * state, ba_t * ba, int bndstart,
34 int start, int end, int fastleak, int slowleak,
37 int a52_downmix_init (int input, int flags, sample_t * level,
38 sample_t clev, sample_t slev);
39 +void downmix_accel_init(uint32_t mm_accel);
40 int a52_downmix_coeff (sample_t * coeff, int acmod, int output, sample_t level,
41 sample_t clev, sample_t slev);
42 -void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
43 +extern void (*a52_downmix) (sample_t * samples, int acmod, int output, sample_t bias,
44 sample_t clev, sample_t slev);
45 -void a52_upmix (sample_t * samples, int acmod, int output);
46 +extern void (*a52_upmix) (sample_t * samples, int acmod, int output);
48 void a52_imdct_init (uint32_t mm_accel);
49 void a52_imdct_256 (sample_t * data, sample_t * delay, sample_t bias);
50 -void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias);
51 +extern void (*a52_imdct_512) (sample_t * data, sample_t * delay, sample_t bias);
52 +void imdct_do_512 (sample_t * data, sample_t * delay, sample_t bias);
53 --- liba52/bitstream.c 2006-06-12 15:05:07.000000000 +0200
54 +++ liba52/bitstream.c 2006-06-05 02:23:02.000000000 +0200
57 #define BUFFER_SIZE 4096
59 +#ifdef ALT_BITSTREAM_READER
63 void a52_bitstream_set_ptr (a52_state_t * state, uint8_t * buf)
67 align = (long)buf & 3;
68 state->buffer_start = (uint32_t *) (buf - align);
70 +#ifdef ALT_BITSTREAM_READER
73 bitstream_get (state, align * 8);
76 --- liba52/bitstream.h 2006-06-12 15:05:07.000000000 +0200
77 +++ liba52/bitstream.h 2006-06-05 02:23:02.000000000 +0200
79 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
82 +/* code from ffmpeg/libavcodec */
83 +#if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC_ == 3 && __GNUC_MINOR__ > 0)
84 +# define always_inline __attribute__((always_inline)) inline
86 +# define always_inline inline
89 +#if defined(__sparc__) || defined(hpux)
91 + * the alt bitstream reader performs unaligned memory accesses; that doesn't work
92 + * on sparc/hpux. For now, disable ALT_BITSTREAM_READER.
94 +#undef ALT_BITSTREAM_READER
96 +// alternative (faster) bitstram reader (reades upto 3 bytes over the end of the input)
97 +#define ALT_BITSTREAM_READER
99 +/* used to avoid missaligned exceptions on some archs (alpha, ...) */
100 +#if defined (ARCH_X86) || defined(ARCH_ARMV4L)
101 +# define unaligned32(a) (*(uint32_t*)(a))
104 +static always_inline uint32_t unaligned32(const void *v) {
107 + } __attribute__((packed));
109 + return ((const struct Unaligned *) v)->i;
111 +# elif defined(__DECC)
112 +static inline uint32_t unaligned32(const void *v) {
113 + return *(const __unaligned uint32_t *) v;
116 +static inline uint32_t unaligned32(const void *v) {
117 + return *(const uint32_t *) v;
124 /* (stolen from the kernel) */
125 #ifdef WORDS_BIGENDIAN
131 -# if 0 && defined (__i386__)
132 +# if defined (__i386__)
134 # define swab32(x) __i386_swab32(x)
135 static inline const uint32_t __i386_swab32(uint32_t x)
141 -((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) | \
142 - (((uint8_t*)&x)[2] << 8) | (((uint8_t*)&x)[3]))
144 +# define swab32(x) __generic_swab32(x)
145 + static always_inline const uint32_t __generic_swab32(uint32_t x)
147 + return ((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) |
148 + (((uint8_t*)&x)[2] << 8) | (((uint8_t*)&x)[3]));
153 +#ifdef ALT_BITSTREAM_READER
157 void a52_bitstream_set_ptr (a52_state_t * state, uint8_t * buf);
158 uint32_t a52_bitstream_get_bh (a52_state_t * state, uint32_t num_bits);
159 int32_t a52_bitstream_get_bh_2 (a52_state_t * state, uint32_t num_bits);
161 static inline uint32_t bitstream_get (a52_state_t * state, uint32_t num_bits)
163 +#ifdef ALT_BITSTREAM_READER
164 + uint32_t result= swab32( unaligned32(((uint8_t *)state->buffer_start)+(indx>>3)) );
166 + result<<= (indx&0x07);
167 + result>>= 32 - num_bits;
174 if (num_bits < state->bits_left) {
178 return a52_bitstream_get_bh (state, num_bits);
182 +static inline void bitstream_skip(a52_state_t * state, int num_bits)
184 +#ifdef ALT_BITSTREAM_READER
187 + bitstream_get(state, num_bits);
191 static inline int32_t bitstream_get_2 (a52_state_t * state, uint32_t num_bits)
193 +#ifdef ALT_BITSTREAM_READER
194 + int32_t result= swab32( unaligned32(((uint8_t *)state->buffer_start)+(indx>>3)) );
196 + result<<= (indx&0x07);
197 + result>>= 32 - num_bits;
204 if (num_bits < state->bits_left) {
208 return a52_bitstream_get_bh_2 (state, num_bits);
211 --- liba52/downmix.c 2006-06-12 15:17:53.000000000 +0200
212 +++ liba52/downmix.c 2006-06-05 02:23:02.000000000 +0200
214 * You should have received a copy of the GNU General Public License
215 * along with this program; if not, write to the Free Software
216 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
218 + * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
224 #include <inttypes.h>
227 #include "a52_internal.h"
228 +#include "mm_accel.h"
230 #define CONVERT(acmod,output) (((output) << 3) + (acmod))
233 +void (*a52_downmix)(sample_t * samples, int acmod, int output, sample_t bias,
234 + sample_t clev, sample_t slev)= NULL;
235 +void (*a52_upmix)(sample_t * samples, int acmod, int output)= NULL;
237 +static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
238 + sample_t clev, sample_t slev);
239 +static void downmix_3dnow (sample_t * samples, int acmod, int output, sample_t bias,
240 + sample_t clev, sample_t slev);
241 +static void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
242 + sample_t clev, sample_t slev);
243 +static void upmix_MMX (sample_t * samples, int acmod, int output);
244 +static void upmix_C (sample_t * samples, int acmod, int output);
246 +void downmix_accel_init(uint32_t mm_accel)
248 + a52_upmix= upmix_C;
249 + a52_downmix= downmix_C;
250 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
251 + if(mm_accel & MM_ACCEL_X86_MMX) a52_upmix= upmix_MMX;
252 + if(mm_accel & MM_ACCEL_X86_SSE) a52_downmix= downmix_SSE;
253 + if(mm_accel & MM_ACCEL_X86_3DNOW) a52_downmix= downmix_3dnow;
257 int a52_downmix_init (int input, int flags, sample_t * level,
258 sample_t clev, sample_t slev)
264 -void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
265 +void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
266 sample_t clev, sample_t slev)
268 switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
272 case CONVERT (A52_3F2R, A52_2F1R):
273 - mix3to2 (samples, bias);
274 + mix3to2 (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
275 move2to1 (samples + 768, samples + 512, bias);
278 @@ -583,12 +615,12 @@
281 case CONVERT (A52_3F1R, A52_3F2R):
282 - memcpy (samples + 1027, samples + 768, 256 * sizeof (sample_t));
283 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
288 -void a52_upmix (sample_t * samples, int acmod, int output)
289 +void upmix_C (sample_t * samples, int acmod, int output)
291 switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
293 @@ -653,3 +685,1137 @@
298 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
299 +static void mix2to1_SSE (sample_t * dest, sample_t * src, sample_t bias)
302 + "movlps %2, %%xmm7 \n\t"
303 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
304 + "mov $-1024, %%"REG_S" \n\t"
307 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
308 + "movaps 16(%0, %%"REG_S"), %%xmm1\n\t"
309 + "addps (%1, %%"REG_S"), %%xmm0 \n\t"
310 + "addps 16(%1, %%"REG_S"), %%xmm1\n\t"
311 + "addps %%xmm7, %%xmm0 \n\t"
312 + "addps %%xmm7, %%xmm1 \n\t"
313 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
314 + "movaps %%xmm1, 16(%1, %%"REG_S")\n\t"
315 + "add $32, %%"REG_S" \n\t"
317 + :: "r" (src+256), "r" (dest+256), "m" (bias)
322 +static void mix3to1_SSE (sample_t * samples, sample_t bias)
325 + "movlps %1, %%xmm7 \n\t"
326 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
327 + "mov $-1024, %%"REG_S" \n\t"
330 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
331 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
332 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
333 + "addps %%xmm7, %%xmm1 \n\t"
334 + "addps %%xmm1, %%xmm0 \n\t"
335 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
336 + "add $16, %%"REG_S" \n\t"
338 + :: "r" (samples+256), "m" (bias)
343 +static void mix4to1_SSE (sample_t * samples, sample_t bias)
346 + "movlps %1, %%xmm7 \n\t"
347 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
348 + "mov $-1024, %%"REG_S" \n\t"
351 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
352 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
353 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
354 + "addps 3072(%0, %%"REG_S"), %%xmm1\n\t"
355 + "addps %%xmm7, %%xmm0 \n\t"
356 + "addps %%xmm1, %%xmm0 \n\t"
357 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
358 + "add $16, %%"REG_S" \n\t"
360 + :: "r" (samples+256), "m" (bias)
365 +static void mix5to1_SSE (sample_t * samples, sample_t bias)
368 + "movlps %1, %%xmm7 \n\t"
369 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
370 + "mov $-1024, %%"REG_S" \n\t"
373 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
374 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
375 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
376 + "addps 3072(%0, %%"REG_S"), %%xmm1\n\t"
377 + "addps %%xmm7, %%xmm0 \n\t"
378 + "addps 4096(%0, %%"REG_S"), %%xmm1\n\t"
379 + "addps %%xmm1, %%xmm0 \n\t"
380 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
381 + "add $16, %%"REG_S" \n\t"
383 + :: "r" (samples+256), "m" (bias)
388 +static void mix3to2_SSE (sample_t * samples, sample_t bias)
391 + "movlps %1, %%xmm7 \n\t"
392 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
393 + "mov $-1024, %%"REG_S" \n\t"
396 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
397 + "addps %%xmm7, %%xmm0 \n\t" //common
398 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
399 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
400 + "addps %%xmm0, %%xmm1 \n\t"
401 + "addps %%xmm0, %%xmm2 \n\t"
402 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
403 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
404 + "add $16, %%"REG_S" \n\t"
406 + :: "r" (samples+256), "m" (bias)
411 +static void mix21to2_SSE (sample_t * left, sample_t * right, sample_t bias)
414 + "movlps %2, %%xmm7 \n\t"
415 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
416 + "mov $-1024, %%"REG_S" \n\t"
419 + "movaps 1024(%1, %%"REG_S"), %%xmm0\n\t"
420 + "addps %%xmm7, %%xmm0 \n\t" //common
421 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
422 + "movaps (%1, %%"REG_S"), %%xmm2 \n\t"
423 + "addps %%xmm0, %%xmm1 \n\t"
424 + "addps %%xmm0, %%xmm2 \n\t"
425 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
426 + "movaps %%xmm2, (%1, %%"REG_S") \n\t"
427 + "add $16, %%"REG_S" \n\t"
429 + :: "r" (left+256), "r" (right+256), "m" (bias)
434 +static void mix21toS_SSE (sample_t * samples, sample_t bias)
437 + "movlps %1, %%xmm7 \n\t"
438 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
439 + "mov $-1024, %%"REG_S" \n\t"
442 + "movaps 2048(%0, %%"REG_S"), %%xmm0\n\t" // surround
443 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
444 + "movaps 1024(%0, %%"REG_S"), %%xmm2\n\t"
445 + "addps %%xmm7, %%xmm1 \n\t"
446 + "addps %%xmm7, %%xmm2 \n\t"
447 + "subps %%xmm0, %%xmm1 \n\t"
448 + "addps %%xmm0, %%xmm2 \n\t"
449 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
450 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
451 + "add $16, %%"REG_S" \n\t"
453 + :: "r" (samples+256), "m" (bias)
458 +static void mix31to2_SSE (sample_t * samples, sample_t bias)
461 + "movlps %1, %%xmm7 \n\t"
462 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
463 + "mov $-1024, %%"REG_S" \n\t"
466 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
467 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t"
468 + "addps %%xmm7, %%xmm0 \n\t" // common
469 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
470 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
471 + "addps %%xmm0, %%xmm1 \n\t"
472 + "addps %%xmm0, %%xmm2 \n\t"
473 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
474 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
475 + "add $16, %%"REG_S" \n\t"
477 + :: "r" (samples+256), "m" (bias)
482 +static void mix31toS_SSE (sample_t * samples, sample_t bias)
485 + "movlps %1, %%xmm7 \n\t"
486 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
487 + "mov $-1024, %%"REG_S" \n\t"
490 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
491 + "movaps 3072(%0, %%"REG_S"), %%xmm3\n\t" // surround
492 + "addps %%xmm7, %%xmm0 \n\t" // common
493 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
494 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
495 + "addps %%xmm0, %%xmm1 \n\t"
496 + "addps %%xmm0, %%xmm2 \n\t"
497 + "subps %%xmm3, %%xmm1 \n\t"
498 + "addps %%xmm3, %%xmm2 \n\t"
499 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
500 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
501 + "add $16, %%"REG_S" \n\t"
503 + :: "r" (samples+256), "m" (bias)
508 +static void mix22toS_SSE (sample_t * samples, sample_t bias)
511 + "movlps %1, %%xmm7 \n\t"
512 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
513 + "mov $-1024, %%"REG_S" \n\t"
516 + "movaps 2048(%0, %%"REG_S"), %%xmm0\n\t"
517 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t" // surround
518 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
519 + "movaps 1024(%0, %%"REG_S"), %%xmm2\n\t"
520 + "addps %%xmm7, %%xmm1 \n\t"
521 + "addps %%xmm7, %%xmm2 \n\t"
522 + "subps %%xmm0, %%xmm1 \n\t"
523 + "addps %%xmm0, %%xmm2 \n\t"
524 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
525 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
526 + "add $16, %%"REG_S" \n\t"
528 + :: "r" (samples+256), "m" (bias)
533 +static void mix32to2_SSE (sample_t * samples, sample_t bias)
536 + "movlps %1, %%xmm7 \n\t"
537 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
538 + "mov $-1024, %%"REG_S" \n\t"
541 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
542 + "addps %%xmm7, %%xmm0 \n\t" // common
543 + "movaps %%xmm0, %%xmm1 \n\t" // common
544 + "addps (%0, %%"REG_S"), %%xmm0 \n\t"
545 + "addps 2048(%0, %%"REG_S"), %%xmm1\n\t"
546 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t"
547 + "addps 4096(%0, %%"REG_S"), %%xmm1\n\t"
548 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
549 + "movaps %%xmm1, 1024(%0, %%"REG_S")\n\t"
550 + "add $16, %%"REG_S" \n\t"
552 + :: "r" (samples+256), "m" (bias)
557 +static void mix32toS_SSE (sample_t * samples, sample_t bias)
560 + "movlps %1, %%xmm7 \n\t"
561 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
562 + "mov $-1024, %%"REG_S" \n\t"
565 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
566 + "movaps 3072(%0, %%"REG_S"), %%xmm2\n\t"
567 + "addps %%xmm7, %%xmm0 \n\t" // common
568 + "addps 4096(%0, %%"REG_S"), %%xmm2\n\t" // surround
569 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
570 + "movaps 2048(%0, %%"REG_S"), %%xmm3\n\t"
571 + "subps %%xmm2, %%xmm1 \n\t"
572 + "addps %%xmm2, %%xmm3 \n\t"
573 + "addps %%xmm0, %%xmm1 \n\t"
574 + "addps %%xmm0, %%xmm3 \n\t"
575 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
576 + "movaps %%xmm3, 1024(%0, %%"REG_S")\n\t"
577 + "add $16, %%"REG_S" \n\t"
579 + :: "r" (samples+256), "m" (bias)
584 +static void move2to1_SSE (sample_t * src, sample_t * dest, sample_t bias)
587 + "movlps %2, %%xmm7 \n\t"
588 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
589 + "mov $-1024, %%"REG_S" \n\t"
592 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
593 + "movaps 16(%0, %%"REG_S"), %%xmm1\n\t"
594 + "addps 1024(%0, %%"REG_S"), %%xmm0\n\t"
595 + "addps 1040(%0, %%"REG_S"), %%xmm1\n\t"
596 + "addps %%xmm7, %%xmm0 \n\t"
597 + "addps %%xmm7, %%xmm1 \n\t"
598 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
599 + "movaps %%xmm1, 16(%1, %%"REG_S")\n\t"
600 + "add $32, %%"REG_S" \n\t"
602 + :: "r" (src+256), "r" (dest+256), "m" (bias)
607 +static void zero_MMX(sample_t * samples)
610 + "mov $-1024, %%"REG_S" \n\t"
611 + "pxor %%mm0, %%mm0 \n\t"
614 + "movq %%mm0, (%0, %%"REG_S") \n\t"
615 + "movq %%mm0, 8(%0, %%"REG_S") \n\t"
616 + "movq %%mm0, 16(%0, %%"REG_S") \n\t"
617 + "movq %%mm0, 24(%0, %%"REG_S") \n\t"
618 + "add $32, %%"REG_S" \n\t"
621 + :: "r" (samples+256)
627 + I hope dest and src will be at least 8 byte aligned and size
628 + will devide on 8 without remain
629 + Note: untested and unused.
631 +static void copy_MMX(void *dest,const void *src,unsigned size)
635 + for(i=0;i<size;i++)
638 + "movq %0, %%mm0\n\t"
639 + "movq 8%0, %%mm1\n\t"
640 + "movq 16%0, %%mm2\n\t"
641 + "movq 24%0, %%mm3\n\t"
642 + "movq 32%0, %%mm4\n\t"
643 + "movq 40%0, %%mm5\n\t"
644 + "movq 48%0, %%mm6\n\t"
645 + "movq 56%0, %%mm7\n\t"
646 + "movq %%mm0, %1\n\t"
647 + "movq %%mm1, 8%1\n\t"
648 + "movq %%mm2, 16%1\n\t"
649 + "movq %%mm3, 24%1\n\t"
650 + "movq %%mm4, 32%1\n\t"
651 + "movq %%mm5, 40%1\n\t"
652 + "movq %%mm6, 48%1\n\t"
653 + "movq %%mm7, 56%1\n\t"
655 + :"m"(src),"m"(dest));
659 +static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
660 + sample_t clev, sample_t slev)
662 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
664 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
665 + memcpy (samples, samples + 256, 256 * sizeof (sample_t));
668 + case CONVERT (A52_CHANNEL, A52_MONO):
669 + case CONVERT (A52_STEREO, A52_MONO):
671 + mix2to1_SSE (samples, samples + 256, bias);
674 + case CONVERT (A52_2F1R, A52_MONO):
677 + case CONVERT (A52_3F, A52_MONO):
679 + mix3to1_SSE (samples, bias);
682 + case CONVERT (A52_3F1R, A52_MONO):
685 + case CONVERT (A52_2F2R, A52_MONO):
688 + mix4to1_SSE (samples, bias);
691 + case CONVERT (A52_3F2R, A52_MONO):
694 + mix5to1_SSE (samples, bias);
697 + case CONVERT (A52_MONO, A52_DOLBY):
698 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
701 + case CONVERT (A52_3F, A52_STEREO):
702 + case CONVERT (A52_3F, A52_DOLBY):
704 + mix3to2_SSE (samples, bias);
707 + case CONVERT (A52_2F1R, A52_STEREO):
710 + mix21to2_SSE (samples, samples + 256, bias);
713 + case CONVERT (A52_2F1R, A52_DOLBY):
714 + mix21toS_SSE (samples, bias);
717 + case CONVERT (A52_3F1R, A52_STEREO):
720 + mix31to2_SSE (samples, bias);
723 + case CONVERT (A52_3F1R, A52_DOLBY):
724 + mix31toS_SSE (samples, bias);
727 + case CONVERT (A52_2F2R, A52_STEREO):
730 + mix2to1_SSE (samples, samples + 512, bias);
731 + mix2to1_SSE (samples + 256, samples + 768, bias);
734 + case CONVERT (A52_2F2R, A52_DOLBY):
735 + mix22toS_SSE (samples, bias);
738 + case CONVERT (A52_3F2R, A52_STEREO):
741 + mix32to2_SSE (samples, bias);
744 + case CONVERT (A52_3F2R, A52_DOLBY):
745 + mix32toS_SSE (samples, bias);
748 + case CONVERT (A52_3F1R, A52_3F):
751 + mix21to2_SSE (samples, samples + 512, bias);
754 + case CONVERT (A52_3F2R, A52_3F):
757 + mix2to1_SSE (samples, samples + 768, bias);
758 + mix2to1_SSE (samples + 512, samples + 1024, bias);
761 + case CONVERT (A52_3F1R, A52_2F1R):
762 + mix3to2_SSE (samples, bias);
763 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
766 + case CONVERT (A52_2F2R, A52_2F1R):
767 + mix2to1_SSE (samples + 512, samples + 768, bias);
770 + case CONVERT (A52_3F2R, A52_2F1R):
771 + mix3to2_SSE (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
772 + move2to1_SSE (samples + 768, samples + 512, bias);
775 + case CONVERT (A52_3F2R, A52_3F1R):
776 + mix2to1_SSE (samples + 768, samples + 1024, bias);
779 + case CONVERT (A52_2F1R, A52_2F2R):
780 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
783 + case CONVERT (A52_3F1R, A52_2F2R):
784 + mix3to2_SSE (samples, bias);
785 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
788 + case CONVERT (A52_3F2R, A52_2F2R):
789 + mix3to2_SSE (samples, bias);
790 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
791 + memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
794 + case CONVERT (A52_3F1R, A52_3F2R):
795 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
800 +static void upmix_MMX (sample_t * samples, int acmod, int output)
802 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
804 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
805 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
808 + case CONVERT (A52_3F2R, A52_MONO):
809 + zero_MMX (samples + 1024);
810 + case CONVERT (A52_3F1R, A52_MONO):
811 + case CONVERT (A52_2F2R, A52_MONO):
812 + zero_MMX (samples + 768);
813 + case CONVERT (A52_3F, A52_MONO):
814 + case CONVERT (A52_2F1R, A52_MONO):
815 + zero_MMX (samples + 512);
816 + case CONVERT (A52_CHANNEL, A52_MONO):
817 + case CONVERT (A52_STEREO, A52_MONO):
818 + zero_MMX (samples + 256);
821 + case CONVERT (A52_3F2R, A52_STEREO):
822 + case CONVERT (A52_3F2R, A52_DOLBY):
823 + zero_MMX (samples + 1024);
824 + case CONVERT (A52_3F1R, A52_STEREO):
825 + case CONVERT (A52_3F1R, A52_DOLBY):
826 + zero_MMX (samples + 768);
827 + case CONVERT (A52_3F, A52_STEREO):
828 + case CONVERT (A52_3F, A52_DOLBY):
830 + memcpy (samples + 512, samples + 256, 256 * sizeof (sample_t));
831 + zero_MMX (samples + 256);
834 + case CONVERT (A52_2F2R, A52_STEREO):
835 + case CONVERT (A52_2F2R, A52_DOLBY):
836 + zero_MMX (samples + 768);
837 + case CONVERT (A52_2F1R, A52_STEREO):
838 + case CONVERT (A52_2F1R, A52_DOLBY):
839 + zero_MMX (samples + 512);
842 + case CONVERT (A52_3F2R, A52_3F):
843 + zero_MMX (samples + 1024);
844 + case CONVERT (A52_3F1R, A52_3F):
845 + case CONVERT (A52_2F2R, A52_2F1R):
846 + zero_MMX (samples + 768);
849 + case CONVERT (A52_3F2R, A52_3F1R):
850 + zero_MMX (samples + 1024);
853 + case CONVERT (A52_3F2R, A52_2F1R):
854 + zero_MMX (samples + 1024);
855 + case CONVERT (A52_3F1R, A52_2F1R):
857 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
860 + case CONVERT (A52_3F2R, A52_2F2R):
861 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
862 + goto mix_31to21_MMX;
866 +static void mix2to1_3dnow (sample_t * dest, sample_t * src, sample_t bias)
869 + "movd %2, %%mm7 \n\t"
870 + "punpckldq %2, %%mm7 \n\t"
871 + "mov $-1024, %%"REG_S" \n\t"
874 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
875 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
876 + "movq 16(%0, %%"REG_S"), %%mm2 \n\t"
877 + "movq 24(%0, %%"REG_S"), %%mm3 \n\t"
878 + "pfadd (%1, %%"REG_S"), %%mm0 \n\t"
879 + "pfadd 8(%1, %%"REG_S"), %%mm1 \n\t"
880 + "pfadd 16(%1, %%"REG_S"), %%mm2 \n\t"
881 + "pfadd 24(%1, %%"REG_S"), %%mm3 \n\t"
882 + "pfadd %%mm7, %%mm0 \n\t"
883 + "pfadd %%mm7, %%mm1 \n\t"
884 + "pfadd %%mm7, %%mm2 \n\t"
885 + "pfadd %%mm7, %%mm3 \n\t"
886 + "movq %%mm0, (%1, %%"REG_S") \n\t"
887 + "movq %%mm1, 8(%1, %%"REG_S") \n\t"
888 + "movq %%mm2, 16(%1, %%"REG_S") \n\t"
889 + "movq %%mm3, 24(%1, %%"REG_S") \n\t"
890 + "add $32, %%"REG_S" \n\t"
892 + :: "r" (src+256), "r" (dest+256), "m" (bias)
897 +static void mix3to1_3dnow (sample_t * samples, sample_t bias)
900 + "movd %1, %%mm7 \n\t"
901 + "punpckldq %1, %%mm7 \n\t"
902 + "mov $-1024, %%"REG_S" \n\t"
905 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
906 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
907 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
908 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
909 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
910 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
911 + "pfadd %%mm7, %%mm0 \n\t"
912 + "pfadd %%mm7, %%mm1 \n\t"
913 + "pfadd %%mm2, %%mm0 \n\t"
914 + "pfadd %%mm3, %%mm1 \n\t"
915 + "movq %%mm0, (%0, %%"REG_S") \n\t"
916 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
917 + "add $16, %%"REG_S" \n\t"
919 + :: "r" (samples+256), "m" (bias)
924 +static void mix4to1_3dnow (sample_t * samples, sample_t bias)
927 + "movd %1, %%mm7 \n\t"
928 + "punpckldq %1, %%mm7 \n\t"
929 + "mov $-1024, %%"REG_S" \n\t"
932 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
933 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
934 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
935 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
936 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
937 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
938 + "pfadd 3072(%0, %%"REG_S"), %%mm2\n\t"
939 + "pfadd 3080(%0, %%"REG_S"), %%mm3\n\t"
940 + "pfadd %%mm7, %%mm0 \n\t"
941 + "pfadd %%mm7, %%mm1 \n\t"
942 + "pfadd %%mm2, %%mm0 \n\t"
943 + "pfadd %%mm3, %%mm1 \n\t"
944 + "movq %%mm0, (%0, %%"REG_S") \n\t"
945 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
946 + "add $16, %%"REG_S" \n\t"
948 + :: "r" (samples+256), "m" (bias)
953 +static void mix5to1_3dnow (sample_t * samples, sample_t bias)
956 + "movd %1, %%mm7 \n\t"
957 + "punpckldq %1, %%mm7 \n\t"
958 + "mov $-1024, %%"REG_S" \n\t"
961 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
962 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
963 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
964 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
965 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
966 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
967 + "pfadd 3072(%0, %%"REG_S"), %%mm2\n\t"
968 + "pfadd 3080(%0, %%"REG_S"), %%mm3\n\t"
969 + "pfadd %%mm7, %%mm0 \n\t"
970 + "pfadd %%mm7, %%mm1 \n\t"
971 + "pfadd 4096(%0, %%"REG_S"), %%mm2\n\t"
972 + "pfadd 4104(%0, %%"REG_S"), %%mm3\n\t"
973 + "pfadd %%mm2, %%mm0 \n\t"
974 + "pfadd %%mm3, %%mm1 \n\t"
975 + "movq %%mm0, (%0, %%"REG_S") \n\t"
976 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
977 + "add $16, %%"REG_S" \n\t"
979 + :: "r" (samples+256), "m" (bias)
984 +static void mix3to2_3dnow (sample_t * samples, sample_t bias)
987 + "movd %1, %%mm7 \n\t"
988 + "punpckldq %1, %%mm7 \n\t"
989 + "mov $-1024, %%"REG_S" \n\t"
992 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
993 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
994 + "pfadd %%mm7, %%mm0 \n\t" //common
995 + "pfadd %%mm7, %%mm1 \n\t" //common
996 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
997 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
998 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
999 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
1000 + "pfadd %%mm0, %%mm2 \n\t"
1001 + "pfadd %%mm1, %%mm3 \n\t"
1002 + "pfadd %%mm0, %%mm4 \n\t"
1003 + "pfadd %%mm1, %%mm5 \n\t"
1004 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1005 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1006 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1007 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1008 + "add $16, %%"REG_S" \n\t"
1010 + :: "r" (samples+256), "m" (bias)
1015 +static void mix21to2_3dnow (sample_t * left, sample_t * right, sample_t bias)
1018 + "movd %2, %%mm7 \n\t"
1019 + "punpckldq %2, %%mm7 \n\t"
1020 + "mov $-1024, %%"REG_S" \n\t"
1023 + "movq 1024(%1, %%"REG_S"), %%mm0\n\t"
1024 + "movq 1032(%1, %%"REG_S"), %%mm1\n\t"
1025 + "pfadd %%mm7, %%mm0 \n\t" //common
1026 + "pfadd %%mm7, %%mm1 \n\t" //common
1027 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1028 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1029 + "movq (%1, %%"REG_S"), %%mm4 \n\t"
1030 + "movq 8(%1, %%"REG_S"), %%mm5 \n\t"
1031 + "pfadd %%mm0, %%mm2 \n\t"
1032 + "pfadd %%mm1, %%mm3 \n\t"
1033 + "pfadd %%mm0, %%mm4 \n\t"
1034 + "pfadd %%mm1, %%mm5 \n\t"
1035 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1036 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1037 + "movq %%mm4, (%1, %%"REG_S") \n\t"
1038 + "movq %%mm5, 8(%1, %%"REG_S") \n\t"
1039 + "add $16, %%"REG_S" \n\t"
1041 + :: "r" (left+256), "r" (right+256), "m" (bias)
1046 +static void mix21toS_3dnow (sample_t * samples, sample_t bias)
1049 + "movd %1, %%mm7 \n\t"
1050 + "punpckldq %1, %%mm7 \n\t"
1051 + "mov $-1024, %%"REG_S" \n\t"
1054 + "movq 2048(%0, %%"REG_S"), %%mm0\n\t" // surround
1055 + "movq 2056(%0, %%"REG_S"), %%mm1\n\t" // surround
1056 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1057 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1058 + "movq 1024(%0, %%"REG_S"), %%mm4\n\t"
1059 + "movq 1032(%0, %%"REG_S"), %%mm5\n\t"
1060 + "pfadd %%mm7, %%mm2 \n\t"
1061 + "pfadd %%mm7, %%mm3 \n\t"
1062 + "pfadd %%mm7, %%mm4 \n\t"
1063 + "pfadd %%mm7, %%mm5 \n\t"
1064 + "pfsub %%mm0, %%mm2 \n\t"
1065 + "pfsub %%mm1, %%mm3 \n\t"
1066 + "pfadd %%mm0, %%mm4 \n\t"
1067 + "pfadd %%mm1, %%mm5 \n\t"
1068 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1069 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1070 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1071 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1072 + "add $16, %%"REG_S" \n\t"
1074 + :: "r" (samples+256), "m" (bias)
1079 +static void mix31to2_3dnow (sample_t * samples, sample_t bias)
1082 + "movd %1, %%mm7 \n\t"
1083 + "punpckldq %1, %%mm7 \n\t"
1084 + "mov $-1024, %%"REG_S" \n\t"
1087 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1088 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1089 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t"
1090 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t"
1091 + "pfadd %%mm7, %%mm0 \n\t" // common
1092 + "pfadd %%mm7, %%mm1 \n\t" // common
1093 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1094 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1095 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
1096 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
1097 + "pfadd %%mm0, %%mm2 \n\t"
1098 + "pfadd %%mm1, %%mm3 \n\t"
1099 + "pfadd %%mm0, %%mm4 \n\t"
1100 + "pfadd %%mm1, %%mm5 \n\t"
1101 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1102 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1103 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1104 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1105 + "add $16, %%"REG_S" \n\t"
1107 + :: "r" (samples+256), "m" (bias)
1112 +static void mix31toS_3dnow (sample_t * samples, sample_t bias)
1115 + "movd %1, %%mm7 \n\t"
1116 + "punpckldq %1, %%mm7 \n\t"
1117 + "mov $-1024, %%"REG_S" \n\t"
1120 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1121 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1122 + "pfadd %%mm7, %%mm0 \n\t" // common
1123 + "pfadd %%mm7, %%mm1 \n\t" // common
1124 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1125 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1126 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
1127 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
1128 + "pfadd %%mm0, %%mm2 \n\t"
1129 + "pfadd %%mm1, %%mm3 \n\t"
1130 + "pfadd %%mm0, %%mm4 \n\t"
1131 + "pfadd %%mm1, %%mm5 \n\t"
1132 + "movq 3072(%0, %%"REG_S"), %%mm0\n\t" // surround
1133 + "movq 3080(%0, %%"REG_S"), %%mm1\n\t" // surround
1134 + "pfsub %%mm0, %%mm2 \n\t"
1135 + "pfsub %%mm1, %%mm3 \n\t"
1136 + "pfadd %%mm0, %%mm4 \n\t"
1137 + "pfadd %%mm1, %%mm5 \n\t"
1138 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1139 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1140 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1141 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1142 + "add $16, %%"REG_S" \n\t"
1144 + :: "r" (samples+256), "m" (bias)
1149 +static void mix22toS_3dnow (sample_t * samples, sample_t bias)
1152 + "movd %1, %%mm7 \n\t"
1153 + "punpckldq %1, %%mm7 \n\t"
1154 + "mov $-1024, %%"REG_S" \n\t"
1157 + "movq 2048(%0, %%"REG_S"), %%mm0\n\t"
1158 + "movq 2056(%0, %%"REG_S"), %%mm1\n\t"
1159 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t" // surround
1160 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t" // surround
1161 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1162 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1163 + "movq 1024(%0, %%"REG_S"), %%mm4\n\t"
1164 + "movq 1032(%0, %%"REG_S"), %%mm5\n\t"
1165 + "pfadd %%mm7, %%mm2 \n\t"
1166 + "pfadd %%mm7, %%mm3 \n\t"
1167 + "pfadd %%mm7, %%mm4 \n\t"
1168 + "pfadd %%mm7, %%mm5 \n\t"
1169 + "pfsub %%mm0, %%mm2 \n\t"
1170 + "pfsub %%mm1, %%mm3 \n\t"
1171 + "pfadd %%mm0, %%mm4 \n\t"
1172 + "pfadd %%mm1, %%mm5 \n\t"
1173 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1174 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1175 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1176 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1177 + "add $16, %%"REG_S" \n\t"
1179 + :: "r" (samples+256), "m" (bias)
1184 +static void mix32to2_3dnow (sample_t * samples, sample_t bias)
1187 + "movd %1, %%mm7 \n\t"
1188 + "punpckldq %1, %%mm7 \n\t"
1189 + "mov $-1024, %%"REG_S" \n\t"
1192 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1193 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1194 + "pfadd %%mm7, %%mm0 \n\t" // common
1195 + "pfadd %%mm7, %%mm1 \n\t" // common
1196 + "movq %%mm0, %%mm2 \n\t" // common
1197 + "movq %%mm1, %%mm3 \n\t" // common
1198 + "pfadd (%0, %%"REG_S"), %%mm0 \n\t"
1199 + "pfadd 8(%0, %%"REG_S"), %%mm1 \n\t"
1200 + "pfadd 2048(%0, %%"REG_S"), %%mm2\n\t"
1201 + "pfadd 2056(%0, %%"REG_S"), %%mm3\n\t"
1202 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t"
1203 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t"
1204 + "pfadd 4096(%0, %%"REG_S"), %%mm2\n\t"
1205 + "pfadd 4104(%0, %%"REG_S"), %%mm3\n\t"
1206 + "movq %%mm0, (%0, %%"REG_S") \n\t"
1207 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
1208 + "movq %%mm2, 1024(%0, %%"REG_S")\n\t"
1209 + "movq %%mm3, 1032(%0, %%"REG_S")\n\t"
1210 + "add $16, %%"REG_S" \n\t"
1212 + :: "r" (samples+256), "m" (bias)
1217 +/* todo: should be optimized better */
1218 +static void mix32toS_3dnow (sample_t * samples, sample_t bias)
1221 + "mov $-1024, %%"REG_S" \n\t"
1224 + "movd %1, %%mm7 \n\t"
1225 + "punpckldq %1, %%mm7 \n\t"
1226 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1227 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1228 + "movq 3072(%0, %%"REG_S"), %%mm4\n\t"
1229 + "movq 3080(%0, %%"REG_S"), %%mm5\n\t"
1230 + "pfadd %%mm7, %%mm0 \n\t" // common
1231 + "pfadd %%mm7, %%mm1 \n\t" // common
1232 + "pfadd 4096(%0, %%"REG_S"), %%mm4\n\t" // surround
1233 + "pfadd 4104(%0, %%"REG_S"), %%mm5\n\t" // surround
1234 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1235 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1236 + "movq 2048(%0, %%"REG_S"), %%mm6\n\t"
1237 + "movq 2056(%0, %%"REG_S"), %%mm7\n\t"
1238 + "pfsub %%mm4, %%mm2 \n\t"
1239 + "pfsub %%mm5, %%mm3 \n\t"
1240 + "pfadd %%mm4, %%mm6 \n\t"
1241 + "pfadd %%mm5, %%mm7 \n\t"
1242 + "pfadd %%mm0, %%mm2 \n\t"
1243 + "pfadd %%mm1, %%mm3 \n\t"
1244 + "pfadd %%mm0, %%mm6 \n\t"
1245 + "pfadd %%mm1, %%mm7 \n\t"
1246 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1247 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1248 + "movq %%mm6, 1024(%0, %%"REG_S")\n\t"
1249 + "movq %%mm7, 1032(%0, %%"REG_S")\n\t"
1250 + "add $16, %%"REG_S" \n\t"
1252 + :: "r" (samples+256), "m" (bias)
1257 +static void move2to1_3dnow (sample_t * src, sample_t * dest, sample_t bias)
1260 + "movd %2, %%mm7 \n\t"
1261 + "punpckldq %2, %%mm7 \n\t"
1262 + "mov $-1024, %%"REG_S" \n\t"
1265 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
1266 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
1267 + "movq 16(%0, %%"REG_S"), %%mm2 \n\t"
1268 + "movq 24(%0, %%"REG_S"), %%mm3 \n\t"
1269 + "pfadd 1024(%0, %%"REG_S"), %%mm0\n\t"
1270 + "pfadd 1032(%0, %%"REG_S"), %%mm1\n\t"
1271 + "pfadd 1040(%0, %%"REG_S"), %%mm2\n\t"
1272 + "pfadd 1048(%0, %%"REG_S"), %%mm3\n\t"
1273 + "pfadd %%mm7, %%mm0 \n\t"
1274 + "pfadd %%mm7, %%mm1 \n\t"
1275 + "pfadd %%mm7, %%mm2 \n\t"
1276 + "pfadd %%mm7, %%mm3 \n\t"
1277 + "movq %%mm0, (%1, %%"REG_S") \n\t"
1278 + "movq %%mm1, 8(%1, %%"REG_S") \n\t"
1279 + "movq %%mm2, 16(%1, %%"REG_S") \n\t"
1280 + "movq %%mm3, 24(%1, %%"REG_S") \n\t"
1281 + "add $32, %%"REG_S" \n\t"
1283 + :: "r" (src+256), "r" (dest+256), "m" (bias)
1288 +static void downmix_3dnow (sample_t * samples, int acmod, int output, sample_t bias,
1289 + sample_t clev, sample_t slev)
1291 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
1293 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
1294 + memcpy (samples, samples + 256, 256 * sizeof (sample_t));
1297 + case CONVERT (A52_CHANNEL, A52_MONO):
1298 + case CONVERT (A52_STEREO, A52_MONO):
1300 + mix2to1_3dnow (samples, samples + 256, bias);
1303 + case CONVERT (A52_2F1R, A52_MONO):
1305 + goto mix_2to1_3dnow;
1306 + case CONVERT (A52_3F, A52_MONO):
1308 + mix3to1_3dnow (samples, bias);
1311 + case CONVERT (A52_3F1R, A52_MONO):
1313 + goto mix_3to1_3dnow;
1314 + case CONVERT (A52_2F2R, A52_MONO):
1316 + goto mix_2to1_3dnow;
1317 + mix4to1_3dnow (samples, bias);
1320 + case CONVERT (A52_3F2R, A52_MONO):
1322 + goto mix_3to1_3dnow;
1323 + mix5to1_3dnow (samples, bias);
1326 + case CONVERT (A52_MONO, A52_DOLBY):
1327 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
1330 + case CONVERT (A52_3F, A52_STEREO):
1331 + case CONVERT (A52_3F, A52_DOLBY):
1333 + mix3to2_3dnow (samples, bias);
1336 + case CONVERT (A52_2F1R, A52_STEREO):
1339 + mix21to2_3dnow (samples, samples + 256, bias);
1342 + case CONVERT (A52_2F1R, A52_DOLBY):
1343 + mix21toS_3dnow (samples, bias);
1346 + case CONVERT (A52_3F1R, A52_STEREO):
1348 + goto mix_3to2_3dnow;
1349 + mix31to2_3dnow (samples, bias);
1352 + case CONVERT (A52_3F1R, A52_DOLBY):
1353 + mix31toS_3dnow (samples, bias);
1356 + case CONVERT (A52_2F2R, A52_STEREO):
1359 + mix2to1_3dnow (samples, samples + 512, bias);
1360 + mix2to1_3dnow (samples + 256, samples + 768, bias);
1363 + case CONVERT (A52_2F2R, A52_DOLBY):
1364 + mix22toS_3dnow (samples, bias);
1367 + case CONVERT (A52_3F2R, A52_STEREO):
1369 + goto mix_3to2_3dnow;
1370 + mix32to2_3dnow (samples, bias);
1373 + case CONVERT (A52_3F2R, A52_DOLBY):
1374 + mix32toS_3dnow (samples, bias);
1377 + case CONVERT (A52_3F1R, A52_3F):
1380 + mix21to2_3dnow (samples, samples + 512, bias);
1383 + case CONVERT (A52_3F2R, A52_3F):
1386 + mix2to1_3dnow (samples, samples + 768, bias);
1387 + mix2to1_3dnow (samples + 512, samples + 1024, bias);
1390 + case CONVERT (A52_3F1R, A52_2F1R):
1391 + mix3to2_3dnow (samples, bias);
1392 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1395 + case CONVERT (A52_2F2R, A52_2F1R):
1396 + mix2to1_3dnow (samples + 512, samples + 768, bias);
1399 + case CONVERT (A52_3F2R, A52_2F1R):
1400 + mix3to2_3dnow (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
1401 + move2to1_3dnow (samples + 768, samples + 512, bias);
1404 + case CONVERT (A52_3F2R, A52_3F1R):
1405 + mix2to1_3dnow (samples + 768, samples + 1024, bias);
1408 + case CONVERT (A52_2F1R, A52_2F2R):
1409 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
1412 + case CONVERT (A52_3F1R, A52_2F2R):
1413 + mix3to2_3dnow (samples, bias);
1414 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1417 + case CONVERT (A52_3F2R, A52_2F2R):
1418 + mix3to2_3dnow (samples, bias);
1419 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1420 + memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
1423 + case CONVERT (A52_3F1R, A52_3F2R):
1424 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
1427 + __asm __volatile("femms":::"memory");
1430 +#endif // ARCH_X86 || ARCH_X86_64
1431 --- liba52/imdct.c 2006-06-12 15:18:27.000000000 +0200
1432 +++ liba52/imdct.c 2006-06-12 19:18:39.000000000 +0200
1434 * You should have received a copy of the GNU General Public License
1435 * along with this program; if not, write to the Free Software
1436 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
1438 + * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
1439 + * 3DNOW optimizations from Nick Kurshev <nickols_k@mail.ru>
1440 + * michael did port them from libac3 (untested, perhaps totally broken)
1441 + * AltiVec optimizations from Romain Dolbeau (romain@dolbeau.org)
1447 #include "a52_internal.h"
1448 #include "mm_accel.h"
1449 +#include "mangle.h"
1451 +void (*a52_imdct_512) (sample_t * data, sample_t * delay, sample_t bias);
1453 +#ifdef RUNTIME_CPUDETECT
1454 +#undef HAVE_3DNOWEX
1457 typedef struct complex_s {
1462 +static const int pm128[128] attribute_used __attribute__((aligned(16))) =
1464 + 0, 16, 32, 48, 64, 80, 96, 112, 8, 40, 72, 104, 24, 56, 88, 120,
1465 + 4, 20, 36, 52, 68, 84, 100, 116, 12, 28, 44, 60, 76, 92, 108, 124,
1466 + 2, 18, 34, 50, 66, 82, 98, 114, 10, 42, 74, 106, 26, 58, 90, 122,
1467 + 6, 22, 38, 54, 70, 86, 102, 118, 14, 46, 78, 110, 30, 62, 94, 126,
1468 + 1, 17, 33, 49, 65, 81, 97, 113, 9, 41, 73, 105, 25, 57, 89, 121,
1469 + 5, 21, 37, 53, 69, 85, 101, 117, 13, 29, 45, 61, 77, 93, 109, 125,
1470 + 3, 19, 35, 51, 67, 83, 99, 115, 11, 43, 75, 107, 27, 59, 91, 123,
1471 + 7, 23, 39, 55, 71, 87, 103, 119, 15, 31, 47, 63, 79, 95, 111, 127
1474 +static uint8_t attribute_used bit_reverse_512[] = {
1475 + 0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70,
1476 + 0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78,
1477 + 0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74,
1478 + 0x0c, 0x4c, 0x2c, 0x6c, 0x1c, 0x5c, 0x3c, 0x7c,
1479 + 0x02, 0x42, 0x22, 0x62, 0x12, 0x52, 0x32, 0x72,
1480 + 0x0a, 0x4a, 0x2a, 0x6a, 0x1a, 0x5a, 0x3a, 0x7a,
1481 + 0x06, 0x46, 0x26, 0x66, 0x16, 0x56, 0x36, 0x76,
1482 + 0x0e, 0x4e, 0x2e, 0x6e, 0x1e, 0x5e, 0x3e, 0x7e,
1483 + 0x01, 0x41, 0x21, 0x61, 0x11, 0x51, 0x31, 0x71,
1484 + 0x09, 0x49, 0x29, 0x69, 0x19, 0x59, 0x39, 0x79,
1485 + 0x05, 0x45, 0x25, 0x65, 0x15, 0x55, 0x35, 0x75,
1486 + 0x0d, 0x4d, 0x2d, 0x6d, 0x1d, 0x5d, 0x3d, 0x7d,
1487 + 0x03, 0x43, 0x23, 0x63, 0x13, 0x53, 0x33, 0x73,
1488 + 0x0b, 0x4b, 0x2b, 0x6b, 0x1b, 0x5b, 0x3b, 0x7b,
1489 + 0x07, 0x47, 0x27, 0x67, 0x17, 0x57, 0x37, 0x77,
1490 + 0x0f, 0x4f, 0x2f, 0x6f, 0x1f, 0x5f, 0x3f, 0x7f};
1492 static uint8_t fftorder[] = {
1493 0,128, 64,192, 32,160,224, 96, 16,144, 80,208,240,112, 48,176,
1494 8,136, 72,200, 40,168,232,104,248,120, 56,184, 24,152,216, 88,
1496 6,134, 70,198, 38,166,230,102,246,118, 54,182, 22,150,214, 86
1499 +static complex_t __attribute__((aligned(16))) buf[128];
1501 +/* Twiddle factor LUT */
1502 +static complex_t __attribute__((aligned(16))) w_1[1];
1503 +static complex_t __attribute__((aligned(16))) w_2[2];
1504 +static complex_t __attribute__((aligned(16))) w_4[4];
1505 +static complex_t __attribute__((aligned(16))) w_8[8];
1506 +static complex_t __attribute__((aligned(16))) w_16[16];
1507 +static complex_t __attribute__((aligned(16))) w_32[32];
1508 +static complex_t __attribute__((aligned(16))) w_64[64];
1509 +static complex_t __attribute__((aligned(16))) * w[7] = {w_1, w_2, w_4, w_8, w_16, w_32, w_64};
1511 +/* Twiddle factors for IMDCT */
1512 +static sample_t __attribute__((aligned(16))) xcos1[128];
1513 +static sample_t __attribute__((aligned(16))) xsin1[128];
1515 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
1516 +// NOTE: SSE needs 16byte alignment or it will segfault
1518 +static float __attribute__((aligned(16))) sseSinCos1c[256];
1519 +static float __attribute__((aligned(16))) sseSinCos1d[256];
1520 +static float attribute_used __attribute__((aligned(16))) ps111_1[4]={1,1,1,-1};
1521 +//static float __attribute__((aligned(16))) sseW0[4];
1522 +static float __attribute__((aligned(16))) sseW1[8];
1523 +static float __attribute__((aligned(16))) sseW2[16];
1524 +static float __attribute__((aligned(16))) sseW3[32];
1525 +static float __attribute__((aligned(16))) sseW4[64];
1526 +static float __attribute__((aligned(16))) sseW5[128];
1527 +static float __attribute__((aligned(16))) sseW6[256];
1528 +static float __attribute__((aligned(16))) *sseW[7]=
1529 + {NULL /*sseW0*/,sseW1,sseW2,sseW3,sseW4,sseW5,sseW6};
1530 +static float __attribute__((aligned(16))) sseWindow[512];
1533 /* Root values for IFFT */
1534 static sample_t roots16[3];
1535 static sample_t roots32[7];
1537 ifft_pass (buf, roots128 - 32, 32);
1540 -void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias)
1541 +void imdct_do_512 (sample_t * data, sample_t * delay, sample_t bias)
1544 sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
1545 @@ -285,6 +365,714 @@
1549 +#ifdef HAVE_ALTIVEC
1552 +#include <altivec.h>
1555 +// used to build registers permutation vectors (vcprm)
1556 +// the 's' are for words in the _s_econd vector
1557 +#define WORD_0 0x00,0x01,0x02,0x03
1558 +#define WORD_1 0x04,0x05,0x06,0x07
1559 +#define WORD_2 0x08,0x09,0x0a,0x0b
1560 +#define WORD_3 0x0c,0x0d,0x0e,0x0f
1561 +#define WORD_s0 0x10,0x11,0x12,0x13
1562 +#define WORD_s1 0x14,0x15,0x16,0x17
1563 +#define WORD_s2 0x18,0x19,0x1a,0x1b
1564 +#define WORD_s3 0x1c,0x1d,0x1e,0x1f
1567 +#define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
1569 +#define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
1572 +// vcprmle is used to keep the same index as in the SSE version.
1573 +// it's the same as vcprm, with the index inversed
1574 +// ('le' is Little Endian)
1575 +#define vcprmle(a,b,c,d) vcprm(d,c,b,a)
1577 +// used to build inverse/identity vectors (vcii)
1578 +// n is _n_egative, p is _p_ositive
1579 +#define FLOAT_n -1.
1583 +#define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
1585 +#define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
1589 +#define FOUROF(a) (a)
1591 +#define FOUROF(a) {a,a,a,a}
1596 +imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
1603 + long two_m_plus_one;
1610 + sample_t *data_ptr;
1611 + sample_t *delay_ptr;
1612 + sample_t *window_ptr;
1614 + /* 512 IMDCT with source and dest data in 'data' */
1616 + /* Pre IFFT complex multiply plus IFFT cmplx conjugate & reordering*/
1617 + for( i=0; i < 128; i++) {
1618 + /* z[i] = (X[256-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */
1619 + int j= bit_reverse_512[i];
1620 + buf[i].real = (data[256-2*j-1] * xcos1[j]) - (data[2*j] * xsin1[j]);
1621 + buf[i].imag = -1.0 * ((data[2*j] * xcos1[j]) + (data[256-2*j-1] * xsin1[j]));
1624 + /* 1. iteration */
1625 + for(i = 0; i < 128; i += 2) {
1627 + tmp_a_r = buf[i].real;
1628 + tmp_a_i = buf[i].imag;
1629 + tmp_b_r = buf[i+1].real;
1630 + tmp_b_i = buf[i+1].imag;
1631 + buf[i].real = tmp_a_r + tmp_b_r;
1632 + buf[i].imag = tmp_a_i + tmp_b_i;
1633 + buf[i+1].real = tmp_a_r - tmp_b_r;
1634 + buf[i+1].imag = tmp_a_i - tmp_b_i;
1636 + vector float temp, bufv;
1638 + bufv = vec_ld(i << 3, (float*)buf);
1639 + temp = vec_perm(bufv, bufv, vcprm(2,3,0,1));
1640 + bufv = vec_madd(bufv, vcii(p,p,n,n), temp);
1641 + vec_st(bufv, i << 3, (float*)buf);
1645 + /* 2. iteration */
1646 + // Note w[1]={{1,0}, {0,-1}}
1647 + for(i = 0; i < 128; i += 4) {
1649 + tmp_a_r = buf[i].real;
1650 + tmp_a_i = buf[i].imag;
1651 + tmp_b_r = buf[i+2].real;
1652 + tmp_b_i = buf[i+2].imag;
1653 + buf[i].real = tmp_a_r + tmp_b_r;
1654 + buf[i].imag = tmp_a_i + tmp_b_i;
1655 + buf[i+2].real = tmp_a_r - tmp_b_r;
1656 + buf[i+2].imag = tmp_a_i - tmp_b_i;
1657 + tmp_a_r = buf[i+1].real;
1658 + tmp_a_i = buf[i+1].imag;
1659 + /* WARNING: im <-> re here ! */
1660 + tmp_b_r = buf[i+3].imag;
1661 + tmp_b_i = buf[i+3].real;
1662 + buf[i+1].real = tmp_a_r + tmp_b_r;
1663 + buf[i+1].imag = tmp_a_i - tmp_b_i;
1664 + buf[i+3].real = tmp_a_r - tmp_b_r;
1665 + buf[i+3].imag = tmp_a_i + tmp_b_i;
1667 + vector float buf01, buf23, temp1, temp2;
1669 + buf01 = vec_ld((i + 0) << 3, (float*)buf);
1670 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1671 + buf23 = vec_perm(buf23,buf23,vcprm(0,1,3,2));
1673 + temp1 = vec_madd(buf23, vcii(p,p,p,n), buf01);
1674 + temp2 = vec_madd(buf23, vcii(n,n,n,p), buf01);
1676 + vec_st(temp1, (i + 0) << 3, (float*)buf);
1677 + vec_st(temp2, (i + 2) << 3, (float*)buf);
1681 + /* 3. iteration */
1682 + for(i = 0; i < 128; i += 8) {
1684 + tmp_a_r = buf[i].real;
1685 + tmp_a_i = buf[i].imag;
1686 + tmp_b_r = buf[i+4].real;
1687 + tmp_b_i = buf[i+4].imag;
1688 + buf[i].real = tmp_a_r + tmp_b_r;
1689 + buf[i].imag = tmp_a_i + tmp_b_i;
1690 + buf[i+4].real = tmp_a_r - tmp_b_r;
1691 + buf[i+4].imag = tmp_a_i - tmp_b_i;
1692 + tmp_a_r = buf[1+i].real;
1693 + tmp_a_i = buf[1+i].imag;
1694 + tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
1695 + tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
1696 + buf[1+i].real = tmp_a_r + tmp_b_r;
1697 + buf[1+i].imag = tmp_a_i + tmp_b_i;
1698 + buf[i+5].real = tmp_a_r - tmp_b_r;
1699 + buf[i+5].imag = tmp_a_i - tmp_b_i;
1700 + tmp_a_r = buf[i+2].real;
1701 + tmp_a_i = buf[i+2].imag;
1702 + /* WARNING re <-> im & sign */
1703 + tmp_b_r = buf[i+6].imag;
1704 + tmp_b_i = - buf[i+6].real;
1705 + buf[i+2].real = tmp_a_r + tmp_b_r;
1706 + buf[i+2].imag = tmp_a_i + tmp_b_i;
1707 + buf[i+6].real = tmp_a_r - tmp_b_r;
1708 + buf[i+6].imag = tmp_a_i - tmp_b_i;
1709 + tmp_a_r = buf[i+3].real;
1710 + tmp_a_i = buf[i+3].imag;
1711 + tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
1712 + tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
1713 + buf[i+3].real = tmp_a_r + tmp_b_r;
1714 + buf[i+3].imag = tmp_a_i + tmp_b_i;
1715 + buf[i+7].real = tmp_a_r - tmp_b_r;
1716 + buf[i+7].imag = tmp_a_i - tmp_b_i;
1718 + vector float buf01, buf23, buf45, buf67;
1720 + buf01 = vec_ld((i + 0) << 3, (float*)buf);
1721 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1723 + tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
1724 + tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
1725 + buf[i+5].real = tmp_b_r;
1726 + buf[i+5].imag = tmp_b_i;
1727 + tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
1728 + tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
1729 + buf[i+7].real = tmp_b_r;
1730 + buf[i+7].imag = tmp_b_i;
1732 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1733 + buf45 = vec_ld((i + 4) << 3, (float*)buf);
1734 + buf67 = vec_ld((i + 6) << 3, (float*)buf);
1735 + buf67 = vec_perm(buf67, buf67, vcprm(1,0,2,3));
1737 + vec_st(vec_add(buf01, buf45), (i + 0) << 3, (float*)buf);
1738 + vec_st(vec_madd(buf67, vcii(p,n,p,p), buf23), (i + 2) << 3, (float*)buf);
1739 + vec_st(vec_sub(buf01, buf45), (i + 4) << 3, (float*)buf);
1740 + vec_st(vec_nmsub(buf67, vcii(p,n,p,p), buf23), (i + 6) << 3, (float*)buf);
1744 + /* 4-7. iterations */
1745 + for (m=3; m < 7; m++) {
1748 + two_m_plus_one = two_m<<1;
1750 + for(i = 0; i < 128; i += two_m_plus_one) {
1751 + for(k = 0; k < two_m; k+=2) {
1754 + int q = p + two_m;
1755 + tmp_a_r = buf[p].real;
1756 + tmp_a_i = buf[p].imag;
1758 + buf[q].real * w[m][k].real -
1759 + buf[q].imag * w[m][k].imag;
1761 + buf[q].imag * w[m][k].real +
1762 + buf[q].real * w[m][k].imag;
1763 + buf[p].real = tmp_a_r + tmp_b_r;
1764 + buf[p].imag = tmp_a_i + tmp_b_i;
1765 + buf[q].real = tmp_a_r - tmp_b_r;
1766 + buf[q].imag = tmp_a_i - tmp_b_i;
1768 + tmp_a_r = buf[(p + 1)].real;
1769 + tmp_a_i = buf[(p + 1)].imag;
1771 + buf[(q + 1)].real * w[m][(k + 1)].real -
1772 + buf[(q + 1)].imag * w[m][(k + 1)].imag;
1774 + buf[(q + 1)].imag * w[m][(k + 1)].real +
1775 + buf[(q + 1)].real * w[m][(k + 1)].imag;
1776 + buf[(p + 1)].real = tmp_a_r + tmp_b_r;
1777 + buf[(p + 1)].imag = tmp_a_i + tmp_b_i;
1778 + buf[(q + 1)].real = tmp_a_r - tmp_b_r;
1779 + buf[(q + 1)].imag = tmp_a_i - tmp_b_i;
1782 + int q = p + two_m;
1783 + vector float vecp, vecq, vecw, temp1, temp2, temp3, temp4;
1784 + const vector float vczero = (const vector float)FOUROF(0.);
1785 + // first compute buf[q] and buf[q+1]
1786 + vecq = vec_ld(q << 3, (float*)buf);
1787 + vecw = vec_ld(0, (float*)&(w[m][k]));
1788 + temp1 = vec_madd(vecq, vecw, vczero);
1789 + temp2 = vec_perm(vecq, vecq, vcprm(1,0,3,2));
1790 + temp2 = vec_madd(temp2, vecw, vczero);
1791 + temp3 = vec_perm(temp1, temp2, vcprm(0,s0,2,s2));
1792 + temp4 = vec_perm(temp1, temp2, vcprm(1,s1,3,s3));
1793 + vecq = vec_madd(temp4, vcii(n,p,n,p), temp3);
1794 + // then butterfly with buf[p] and buf[p+1]
1795 + vecp = vec_ld(p << 3, (float*)buf);
1797 + temp1 = vec_add(vecp, vecq);
1798 + temp2 = vec_sub(vecp, vecq);
1800 + vec_st(temp1, p << 3, (float*)buf);
1801 + vec_st(temp2, q << 3, (float*)buf);
1807 + /* Post IFFT complex multiply plus IFFT complex conjugate*/
1808 + for( i=0; i < 128; i+=4) {
1809 + /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
1811 + tmp_a_r = buf[(i + 0)].real;
1812 + tmp_a_i = -1.0 * buf[(i + 0)].imag;
1813 + buf[(i + 0)].real =
1814 + (tmp_a_r * xcos1[(i + 0)]) - (tmp_a_i * xsin1[(i + 0)]);
1815 + buf[(i + 0)].imag =
1816 + (tmp_a_r * xsin1[(i + 0)]) + (tmp_a_i * xcos1[(i + 0)]);
1818 + tmp_a_r = buf[(i + 1)].real;
1819 + tmp_a_i = -1.0 * buf[(i + 1)].imag;
1820 + buf[(i + 1)].real =
1821 + (tmp_a_r * xcos1[(i + 1)]) - (tmp_a_i * xsin1[(i + 1)]);
1822 + buf[(i + 1)].imag =
1823 + (tmp_a_r * xsin1[(i + 1)]) + (tmp_a_i * xcos1[(i + 1)]);
1825 + tmp_a_r = buf[(i + 2)].real;
1826 + tmp_a_i = -1.0 * buf[(i + 2)].imag;
1827 + buf[(i + 2)].real =
1828 + (tmp_a_r * xcos1[(i + 2)]) - (tmp_a_i * xsin1[(i + 2)]);
1829 + buf[(i + 2)].imag =
1830 + (tmp_a_r * xsin1[(i + 2)]) + (tmp_a_i * xcos1[(i + 2)]);
1832 + tmp_a_r = buf[(i + 3)].real;
1833 + tmp_a_i = -1.0 * buf[(i + 3)].imag;
1834 + buf[(i + 3)].real =
1835 + (tmp_a_r * xcos1[(i + 3)]) - (tmp_a_i * xsin1[(i + 3)]);
1836 + buf[(i + 3)].imag =
1837 + (tmp_a_r * xsin1[(i + 3)]) + (tmp_a_i * xcos1[(i + 3)]);
1839 + vector float bufv_0, bufv_2, cosv, sinv, temp1, temp2;
1840 + vector float temp0022, temp1133, tempCS01;
1841 + const vector float vczero = (const vector float)FOUROF(0.);
1843 + bufv_0 = vec_ld((i + 0) << 3, (float*)buf);
1844 + bufv_2 = vec_ld((i + 2) << 3, (float*)buf);
1846 + cosv = vec_ld(i << 2, xcos1);
1847 + sinv = vec_ld(i << 2, xsin1);
1849 + temp0022 = vec_perm(bufv_0, bufv_0, vcprm(0,0,2,2));
1850 + temp1133 = vec_perm(bufv_0, bufv_0, vcprm(1,1,3,3));
1851 + tempCS01 = vec_perm(cosv, sinv, vcprm(0,s0,1,s1));
1852 + temp1 = vec_madd(temp0022, tempCS01, vczero);
1853 + tempCS01 = vec_perm(cosv, sinv, vcprm(s0,0,s1,1));
1854 + temp2 = vec_madd(temp1133, tempCS01, vczero);
1855 + bufv_0 = vec_madd(temp2, vcii(p,n,p,n), temp1);
1857 + vec_st(bufv_0, (i + 0) << 3, (float*)buf);
1859 + /* idem with bufv_2 and high-order cosv/sinv */
1861 + temp0022 = vec_perm(bufv_2, bufv_2, vcprm(0,0,2,2));
1862 + temp1133 = vec_perm(bufv_2, bufv_2, vcprm(1,1,3,3));
1863 + tempCS01 = vec_perm(cosv, sinv, vcprm(2,s2,3,s3));
1864 + temp1 = vec_madd(temp0022, tempCS01, vczero);
1865 + tempCS01 = vec_perm(cosv, sinv, vcprm(s2,2,s3,3));
1866 + temp2 = vec_madd(temp1133, tempCS01, vczero);
1867 + bufv_2 = vec_madd(temp2, vcii(p,n,p,n), temp1);
1869 + vec_st(bufv_2, (i + 2) << 3, (float*)buf);
1875 + delay_ptr = delay;
1876 + window_ptr = a52_imdct_window;
1878 + /* Window and convert to real valued signal */
1879 + for(i=0; i< 64; i++) {
1880 + *data_ptr++ = -buf[64+i].imag * *window_ptr++ + *delay_ptr++ + bias;
1881 + *data_ptr++ = buf[64-i-1].real * *window_ptr++ + *delay_ptr++ + bias;
1884 + for(i=0; i< 64; i++) {
1885 + *data_ptr++ = -buf[i].real * *window_ptr++ + *delay_ptr++ + bias;
1886 + *data_ptr++ = buf[128-i-1].imag * *window_ptr++ + *delay_ptr++ + bias;
1889 + /* The trailing edge of the window goes into the delay line */
1890 + delay_ptr = delay;
1892 + for(i=0; i< 64; i++) {
1893 + *delay_ptr++ = -buf[64+i].real * *--window_ptr;
1894 + *delay_ptr++ = buf[64-i-1].imag * *--window_ptr;
1897 + for(i=0; i<64; i++) {
1898 + *delay_ptr++ = buf[i].imag * *--window_ptr;
1899 + *delay_ptr++ = -buf[128-i-1].real * *--window_ptr;
1905 +// Stuff below this line is borrowed from libac3
1906 +#include "srfftp.h"
1907 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
1909 +#define HAVE_3DNOW 1
1911 +#include "srfftp_3dnow.h"
1913 +const i_cmplx_t x_plus_minus_3dnow __attribute__ ((aligned (8))) = {{ 0x00000000UL, 0x80000000UL }};
1914 +const i_cmplx_t x_minus_plus_3dnow __attribute__ ((aligned (8))) = {{ 0x80000000UL, 0x00000000UL }};
1915 +const complex_t HSQRT2_3DNOW __attribute__ ((aligned (8))) = { 0.707106781188, 0.707106781188 };
1917 +#undef HAVE_3DNOWEX
1918 +#include "imdct_3dnow.h"
1919 +#define HAVE_3DNOWEX
1920 +#include "imdct_3dnow.h"
1923 +imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
1929 + long two_m_plus_one;
1930 + long two_m_plus_one_shl3;
1931 + complex_t *buf_offset;
1933 +/* sample_t tmp_a_i;
1936 + sample_t tmp_b_r;*/
1938 + sample_t *data_ptr;
1939 + sample_t *delay_ptr;
1940 + sample_t *window_ptr;
1942 + /* 512 IMDCT with source and dest data in 'data' */
1943 + /* see the c version (dct_do_512()), its allmost identical, just in C */
1945 + /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
1946 + /* Bit reversed shuffling */
1948 + "xor %%"REG_S", %%"REG_S" \n\t"
1949 + "lea "MANGLE(bit_reverse_512)", %%"REG_a"\n\t"
1950 + "mov $1008, %%"REG_D" \n\t"
1951 + "push %%"REG_BP" \n\t" //use ebp without telling gcc
1954 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // XXXI
1955 + "movhps 8(%0, %%"REG_D"), %%xmm0 \n\t" // RXXI
1956 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // XXXi
1957 + "movhps (%0, %%"REG_D"), %%xmm1 \n\t" // rXXi
1958 + "shufps $0x33, %%xmm1, %%xmm0 \n\t" // irIR
1959 + "movaps "MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm2\n\t"
1960 + "mulps %%xmm0, %%xmm2 \n\t"
1961 + "shufps $0xB1, %%xmm0, %%xmm0 \n\t" // riRI
1962 + "mulps "MANGLE(sseSinCos1d)"(%%"REG_S"), %%xmm0\n\t"
1963 + "subps %%xmm0, %%xmm2 \n\t"
1964 + "movzb (%%"REG_a"), %%"REG_d" \n\t"
1965 + "movzb 1(%%"REG_a"), %%"REG_BP" \n\t"
1966 + "movlps %%xmm2, (%1, %%"REG_d", 8) \n\t"
1967 + "movhps %%xmm2, (%1, %%"REG_BP", 8) \n\t"
1968 + "add $16, %%"REG_S" \n\t"
1969 + "add $2, %%"REG_a" \n\t" // avoid complex addressing for P4 crap
1970 + "sub $16, %%"REG_D" \n\t"
1972 + "pop %%"REG_BP" \n\t"//no we didnt touch ebp *g*
1973 + :: "b" (data), "c" (buf)
1974 + : "%"REG_S, "%"REG_D, "%"REG_a, "%"REG_d
1979 +/* unoptimized variant
1980 + for (m=1; m < 7; m++) {
1986 + two_m_plus_one = (1 << (m+1));
1988 + for(i = 0; i < 128; i += two_m_plus_one) {
1989 + for(k = 0; k < two_m; k++) {
1992 + tmp_a_r = buf[p].real;
1993 + tmp_a_i = buf[p].imag;
1994 + tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
1995 + tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
1996 + buf[p].real = tmp_a_r + tmp_b_r;
1997 + buf[p].imag = tmp_a_i + tmp_b_i;
1998 + buf[q].real = tmp_a_r - tmp_b_r;
1999 + buf[q].imag = tmp_a_i - tmp_b_i;
2005 + /* 1. iteration */
2006 + // Note w[0][0]={1,0}
2008 + "xorps %%xmm1, %%xmm1 \n\t"
2009 + "xorps %%xmm2, %%xmm2 \n\t"
2010 + "mov %0, %%"REG_S" \n\t"
2013 + "movlps (%%"REG_S"), %%xmm0\n\t" //buf[p]
2014 + "movlps 8(%%"REG_S"), %%xmm1\n\t" //buf[q]
2015 + "movhps (%%"REG_S"), %%xmm0\n\t" //buf[p]
2016 + "movhps 8(%%"REG_S"), %%xmm2\n\t" //buf[q]
2017 + "addps %%xmm1, %%xmm0 \n\t"
2018 + "subps %%xmm2, %%xmm0 \n\t"
2019 + "movaps %%xmm0, (%%"REG_S")\n\t"
2020 + "add $16, %%"REG_S" \n\t"
2021 + "cmp %1, %%"REG_S" \n\t"
2023 + :: "g" (buf), "r" (buf + 128)
2027 + /* 2. iteration */
2028 + // Note w[1]={{1,0}, {0,-1}}
2030 + "movaps "MANGLE(ps111_1)", %%xmm7\n\t" // 1,1,1,-1
2031 + "mov %0, %%"REG_S" \n\t"
2034 + "movaps 16(%%"REG_S"), %%xmm2 \n\t" //r2,i2,r3,i3
2035 + "shufps $0xB4, %%xmm2, %%xmm2 \n\t" //r2,i2,i3,r3
2036 + "mulps %%xmm7, %%xmm2 \n\t" //r2,i2,i3,-r3
2037 + "movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
2038 + "movaps (%%"REG_S"), %%xmm1 \n\t" //r0,i0,r1,i1
2039 + "addps %%xmm2, %%xmm0 \n\t"
2040 + "subps %%xmm2, %%xmm1 \n\t"
2041 + "movaps %%xmm0, (%%"REG_S") \n\t"
2042 + "movaps %%xmm1, 16(%%"REG_S") \n\t"
2043 + "add $32, %%"REG_S" \n\t"
2044 + "cmp %1, %%"REG_S" \n\t"
2046 + :: "g" (buf), "r" (buf + 128)
2050 + /* 3. iteration */
2052 + Note sseW2+0={1,1,sqrt(2),sqrt(2))
2053 + Note sseW2+16={0,0,sqrt(2),-sqrt(2))
2054 + Note sseW2+32={0,0,-sqrt(2),-sqrt(2))
2055 + Note sseW2+48={1,-1,sqrt(2),-sqrt(2))
2058 + "movaps 48+"MANGLE(sseW2)", %%xmm6\n\t"
2059 + "movaps 16+"MANGLE(sseW2)", %%xmm7\n\t"
2060 + "xorps %%xmm5, %%xmm5 \n\t"
2061 + "xorps %%xmm2, %%xmm2 \n\t"
2062 + "mov %0, %%"REG_S" \n\t"
2065 + "movaps 32(%%"REG_S"), %%xmm2 \n\t" //r4,i4,r5,i5
2066 + "movaps 48(%%"REG_S"), %%xmm3 \n\t" //r6,i6,r7,i7
2067 + "movaps "MANGLE(sseW2)", %%xmm4 \n\t" //r4,i4,r5,i5
2068 + "movaps 32+"MANGLE(sseW2)", %%xmm5\n\t" //r6,i6,r7,i7
2069 + "mulps %%xmm2, %%xmm4 \n\t"
2070 + "mulps %%xmm3, %%xmm5 \n\t"
2071 + "shufps $0xB1, %%xmm2, %%xmm2 \n\t" //i4,r4,i5,r5
2072 + "shufps $0xB1, %%xmm3, %%xmm3 \n\t" //i6,r6,i7,r7
2073 + "mulps %%xmm6, %%xmm3 \n\t"
2074 + "mulps %%xmm7, %%xmm2 \n\t"
2075 + "movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
2076 + "movaps 16(%%"REG_S"), %%xmm1 \n\t" //r2,i2,r3,i3
2077 + "addps %%xmm4, %%xmm2 \n\t"
2078 + "addps %%xmm5, %%xmm3 \n\t"
2079 + "movaps %%xmm2, %%xmm4 \n\t"
2080 + "movaps %%xmm3, %%xmm5 \n\t"
2081 + "addps %%xmm0, %%xmm2 \n\t"
2082 + "addps %%xmm1, %%xmm3 \n\t"
2083 + "subps %%xmm4, %%xmm0 \n\t"
2084 + "subps %%xmm5, %%xmm1 \n\t"
2085 + "movaps %%xmm2, (%%"REG_S") \n\t"
2086 + "movaps %%xmm3, 16(%%"REG_S") \n\t"
2087 + "movaps %%xmm0, 32(%%"REG_S") \n\t"
2088 + "movaps %%xmm1, 48(%%"REG_S") \n\t"
2089 + "add $64, %%"REG_S" \n\t"
2090 + "cmp %1, %%"REG_S" \n\t"
2092 + :: "g" (buf), "r" (buf + 128)
2096 + /* 4-7. iterations */
2097 + for (m=3; m < 7; m++) {
2099 + two_m_plus_one = two_m<<1;
2100 + two_m_plus_one_shl3 = (two_m_plus_one<<3);
2101 + buf_offset = buf+128;
2103 + "mov %0, %%"REG_S" \n\t"
2106 + "xor %%"REG_D", %%"REG_D" \n\t" // k
2107 + "lea (%%"REG_S", %3), %%"REG_d" \n\t"
2109 + "movaps (%%"REG_d", %%"REG_D"), %%xmm1 \n\t"
2110 + "movaps (%4, %%"REG_D", 2), %%xmm2 \n\t"
2111 + "mulps %%xmm1, %%xmm2 \n\t"
2112 + "shufps $0xB1, %%xmm1, %%xmm1 \n\t"
2113 + "mulps 16(%4, %%"REG_D", 2), %%xmm1 \n\t"
2114 + "movaps (%%"REG_S", %%"REG_D"), %%xmm0 \n\t"
2115 + "addps %%xmm2, %%xmm1 \n\t"
2116 + "movaps %%xmm1, %%xmm2 \n\t"
2117 + "addps %%xmm0, %%xmm1 \n\t"
2118 + "subps %%xmm2, %%xmm0 \n\t"
2119 + "movaps %%xmm1, (%%"REG_S", %%"REG_D") \n\t"
2120 + "movaps %%xmm0, (%%"REG_d", %%"REG_D") \n\t"
2121 + "add $16, %%"REG_D" \n\t"
2122 + "cmp %3, %%"REG_D" \n\t" //FIXME (opt) count against 0
2124 + "add %2, %%"REG_S" \n\t"
2125 + "cmp %1, %%"REG_S" \n\t"
2127 + :: "g" (buf), "m" (buf_offset), "m" (two_m_plus_one_shl3), "r" (two_m<<3),
2129 + : "%"REG_S, "%"REG_D, "%"REG_d
2133 + /* Post IFFT complex multiply plus IFFT complex conjugate*/
2135 + "mov $-1024, %%"REG_S" \n\t"
2138 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
2139 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
2140 + "shufps $0xB1, %%xmm0, %%xmm0 \n\t"
2141 + "mulps 1024+"MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm1\n\t"
2142 + "mulps 1024+"MANGLE(sseSinCos1d)"(%%"REG_S"), %%xmm0\n\t"
2143 + "addps %%xmm1, %%xmm0 \n\t"
2144 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
2145 + "add $16, %%"REG_S" \n\t"
2153 + delay_ptr = delay;
2154 + window_ptr = a52_imdct_window;
2156 + /* Window and convert to real valued signal */
2158 + "xor %%"REG_D", %%"REG_D" \n\t" // 0
2159 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2160 + "movss %3, %%xmm2 \n\t" // bias
2161 + "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
2164 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
2165 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
2166 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
2167 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // ? B A ?
2168 + "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
2169 + "mulps "MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2170 + "addps (%2, %%"REG_S"), %%xmm0 \n\t"
2171 + "addps %%xmm2, %%xmm0 \n\t"
2172 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2173 + "add $16, %%"REG_S" \n\t"
2174 + "sub $16, %%"REG_D" \n\t"
2175 + "cmp $512, %%"REG_S" \n\t"
2177 + :: "r" (buf+64), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
2178 + : "%"REG_S, "%"REG_D
2182 +// window_ptr+=128;
2185 + "mov $1024, %%"REG_D" \n\t" // 512
2186 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2187 + "movss %3, %%xmm2 \n\t" // bias
2188 + "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
2191 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
2192 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
2193 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
2194 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // B ? ? A
2195 + "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
2196 + "mulps 512+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2197 + "addps (%2, %%"REG_S"), %%xmm0 \n\t"
2198 + "addps %%xmm2, %%xmm0 \n\t"
2199 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2200 + "add $16, %%"REG_S" \n\t"
2201 + "sub $16, %%"REG_D" \n\t"
2202 + "cmp $512, %%"REG_S" \n\t"
2204 + :: "r" (buf), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
2205 + : "%"REG_S, "%"REG_D
2208 +// window_ptr+=128;
2210 + /* The trailing edge of the window goes into the delay line */
2211 + delay_ptr = delay;
2214 + "xor %%"REG_D", %%"REG_D" \n\t" // 0
2215 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2218 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
2219 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
2220 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
2221 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // B ? ? A
2222 + "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
2223 + "mulps 1024+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2224 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2225 + "add $16, %%"REG_S" \n\t"
2226 + "sub $16, %%"REG_D" \n\t"
2227 + "cmp $512, %%"REG_S" \n\t"
2229 + :: "r" (buf+64), "r" (delay_ptr)
2230 + : "%"REG_S, "%"REG_D
2233 +// window_ptr-=128;
2236 + "mov $1024, %%"REG_D" \n\t" // 1024
2237 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2240 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
2241 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
2242 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
2243 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // ? B A ?
2244 + "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
2245 + "mulps 1536+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2246 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2247 + "add $16, %%"REG_S" \n\t"
2248 + "sub $16, %%"REG_D" \n\t"
2249 + "cmp $512, %%"REG_S" \n\t"
2251 + :: "r" (buf), "r" (delay_ptr)
2252 + : "%"REG_S, "%"REG_D
2255 +#endif // ARCH_X86 || ARCH_X86_64
2257 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
2260 @@ -364,7 +1152,7 @@
2262 void a52_imdct_init (uint32_t mm_accel)
2268 /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
2269 @@ -416,6 +1204,99 @@
2270 post2[i].real = cos ((M_PI / 128) * (i + 0.5));
2271 post2[i].imag = sin ((M_PI / 128) * (i + 0.5));
2273 + for (i = 0; i < 128; i++) {
2274 + xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));
2275 + xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
2277 + for (i = 0; i < 7; i++) {
2279 + for (k = 0; k < j; k++) {
2280 + w[i][k].real = cos (-M_PI * k / j);
2281 + w[i][k].imag = sin (-M_PI * k / j);
2284 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
2285 + for (i = 0; i < 128; i++) {
2286 + sseSinCos1c[2*i+0]= xcos1[i];
2287 + sseSinCos1c[2*i+1]= -xcos1[i];
2288 + sseSinCos1d[2*i+0]= xsin1[i];
2289 + sseSinCos1d[2*i+1]= xsin1[i];
2291 + for (i = 1; i < 7; i++) {
2293 + for (k = 0; k < j; k+=2) {
2295 + sseW[i][4*k + 0] = w[i][k+0].real;
2296 + sseW[i][4*k + 1] = w[i][k+0].real;
2297 + sseW[i][4*k + 2] = w[i][k+1].real;
2298 + sseW[i][4*k + 3] = w[i][k+1].real;
2300 + sseW[i][4*k + 4] = -w[i][k+0].imag;
2301 + sseW[i][4*k + 5] = w[i][k+0].imag;
2302 + sseW[i][4*k + 6] = -w[i][k+1].imag;
2303 + sseW[i][4*k + 7] = w[i][k+1].imag;
2305 + //we multiply more or less uninitalized numbers so we need to use exactly 0.0
2308 +// sseW[i][4*k + 0]= sseW[i][4*k + 1]= 1.0;
2309 + sseW[i][4*k + 4]= sseW[i][4*k + 5]= 0.0;
2314 + sseW[i][4*k + 0]= sseW[i][4*k + 1]= 0.0;
2315 +// sseW[i][4*k + 4]= -(sseW[i][4*k + 5]= -1.0);
2320 + for(i=0; i<128; i++)
2322 + sseWindow[2*i+0]= -a52_imdct_window[2*i+0];
2323 + sseWindow[2*i+1]= a52_imdct_window[2*i+1];
2326 + for(i=0; i<64; i++)
2328 + sseWindow[256 + 2*i+0]= -a52_imdct_window[254 - 2*i+1];
2329 + sseWindow[256 + 2*i+1]= a52_imdct_window[254 - 2*i+0];
2330 + sseWindow[384 + 2*i+0]= a52_imdct_window[126 - 2*i+1];
2331 + sseWindow[384 + 2*i+1]= -a52_imdct_window[126 - 2*i+0];
2334 + a52_imdct_512 = imdct_do_512;
2335 + ifft128 = ifft128_c;
2336 + ifft64 = ifft64_c;
2338 +#if defined(ARCH_X86) || defined(ARCH_X86_64)
2339 + if(mm_accel & MM_ACCEL_X86_SSE)
2341 + fprintf (stderr, "Using SSE optimized IMDCT transform\n");
2342 + a52_imdct_512 = imdct_do_512_sse;
2345 + if(mm_accel & MM_ACCEL_X86_3DNOWEXT)
2347 + fprintf (stderr, "Using 3DNowEx optimized IMDCT transform\n");
2348 + a52_imdct_512 = imdct_do_512_3dnowex;
2351 + if(mm_accel & MM_ACCEL_X86_3DNOW)
2353 + fprintf (stderr, "Using 3DNow optimized IMDCT transform\n");
2354 + a52_imdct_512 = imdct_do_512_3dnow;
2357 +#endif // ARCH_X86 || ARCH_X86_64
2358 +#ifdef HAVE_ALTIVEC
2359 + if (mm_accel & MM_ACCEL_PPC_ALTIVEC)
2361 + fprintf(stderr, "Using AltiVec optimized IMDCT transform\n");
2362 + a52_imdct_512 = imdct_do_512_altivec;
2367 #ifdef LIBA52_DJBFFT
2368 if (mm_accel & MM_ACCEL_DJBFFT) {
2369 @@ -426,7 +1307,5 @@
2372 fprintf (stderr, "No accelerated IMDCT transform found\n");
2373 - ifft128 = ifft128_c;
2374 - ifft64 = ifft64_c;
2377 --- include/mm_accel.h 2006-06-12 15:05:00.000000000 +0200
2378 +++ liba52/mm_accel.h 2006-06-05 02:23:04.000000000 +0200
2380 /* x86 accelerations */
2381 #define MM_ACCEL_X86_MMX 0x80000000
2382 #define MM_ACCEL_X86_3DNOW 0x40000000
2383 +#define MM_ACCEL_X86_3DNOWEXT 0x08000000
2384 #define MM_ACCEL_X86_MMXEXT 0x20000000
2385 +#define MM_ACCEL_X86_SSE 0x10000000
2387 +/* PPC accelerations */
2388 +#define MM_ACCEL_PPC_ALTIVEC 0x00010000
2390 uint32_t mm_accel (void);
2392 --- liba52/parse.c 2006-06-12 15:05:07.000000000 +0200
2393 +++ liba52/parse.c 2006-06-12 14:51:33.000000000 +0200
2400 #include <inttypes.h>
2403 #include "a52_internal.h"
2404 #include "bitstream.h"
2406 +#include "mm_accel.h"
2408 #ifdef HAVE_MEMALIGN
2409 /* some systems have memalign() but no declaration for it */
2410 void * memalign (size_t align, size_t size);
2412 -/* assume malloc alignment is sufficient */
2413 -#define memalign(align,size) malloc (size)
2420 state->samples = memalign (16, 256 * 12 * sizeof (sample_t));
2421 +#if defined(__MINGW32__) && defined(HAVE_SSE)
2422 + for(i=0;i<10;i++){
2423 + if((int)state->samples%16){
2424 + sample_t* samplestmp=malloc(256 * 12 * sizeof (sample_t));
2425 + free(state->samples);
2426 + state->samples = samplestmp;
2431 + if(((int)state->samples%16) && (mm_accel&MM_ACCEL_X86_SSE)){
2432 + mm_accel &=~MM_ACCEL_X86_SSE;
2433 + fprintf(stderr, "liba52: unable to get 16 byte aligned memory disabling usage of SSE instructions\n");
2436 if (state->samples == NULL) {
2440 state->lfsr_state = 1;
2442 a52_imdct_init (mm_accel);
2443 + downmix_accel_init(mm_accel);
2448 state->acmod = acmod = buf[6] >> 5;
2450 a52_bitstream_set_ptr (state, buf + 6);
2451 - bitstream_get (state, 3); /* skip acmod we already parsed */
2452 + bitstream_skip (state, 3); /* skip acmod we already parsed */
2454 if ((acmod == 2) && (bitstream_get (state, 2) == 2)) /* dsurmod */
2456 @@ -172,28 +191,28 @@
2460 - bitstream_get (state, 5); /* dialnorm */
2461 + bitstream_skip (state, 5); /* dialnorm */
2462 if (bitstream_get (state, 1)) /* compre */
2463 - bitstream_get (state, 8); /* compr */
2464 + bitstream_skip (state, 8); /* compr */
2465 if (bitstream_get (state, 1)) /* langcode */
2466 - bitstream_get (state, 8); /* langcod */
2467 + bitstream_skip (state, 8); /* langcod */
2468 if (bitstream_get (state, 1)) /* audprodie */
2469 - bitstream_get (state, 7); /* mixlevel + roomtyp */
2470 + bitstream_skip (state, 7); /* mixlevel + roomtyp */
2471 } while (chaninfo--);
2473 - bitstream_get (state, 2); /* copyrightb + origbs */
2474 + bitstream_skip (state, 2); /* copyrightb + origbs */
2476 if (bitstream_get (state, 1)) /* timecod1e */
2477 - bitstream_get (state, 14); /* timecod1 */
2478 + bitstream_skip (state, 14); /* timecod1 */
2479 if (bitstream_get (state, 1)) /* timecod2e */
2480 - bitstream_get (state, 14); /* timecod2 */
2481 + bitstream_skip (state, 14); /* timecod2 */
2483 if (bitstream_get (state, 1)) { /* addbsie */
2486 addbsil = bitstream_get (state, 6);
2488 - bitstream_get (state, 8); /* addbsi */
2489 + bitstream_skip (state, 8); /* addbsi */
2490 } while (addbsil--);
2494 state->fbw_expbap[i].exp[0],
2495 state->fbw_expbap[i].exp + 1))
2497 - bitstream_get (state, 2); /* gainrng */
2498 + bitstream_skip (state, 2); /* gainrng */
2500 if (lfeexpstr != EXP_REUSE) {
2503 if (bitstream_get (state, 1)) { /* skiple */
2504 i = bitstream_get (state, 9); /* skipl */
2506 - bitstream_get (state, 8);
2507 + bitstream_skip (state, 8);
2510 samples = state->samples;