| blob | 1b3574f0432f9eddc625be3093eaa69ecfa8b286 |
1 --- include/a52.h 2006-06-12 15:04:57.000000000 +0200
2 +++ liba52/a52.h 2006-06-05 02:23:02.000000000 +0200
3 @@ -59,4 +66,9 @@
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);
9 +
10 +uint16_t crc16_block(uint8_t *data,uint32_t num_bytes);
11 +
12 #endif /* A52_H */
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
15 @@ -103,18 +107,34 @@
16 #define DELTA_BIT_NONE (2)
17 #define DELTA_BIT_RESERVED (3)
19 +#if ARCH_X86_64
20 +# define REG_a "rax"
21 +# define REG_d "rdx"
22 +# define REG_S "rsi"
23 +# define REG_D "rdi"
24 +# define REG_BP "rbp"
25 +#else
26 +# define REG_a "eax"
27 +# define REG_d "edx"
28 +# define REG_S "esi"
29 +# define REG_D "edi"
30 +# define REG_BP "ebp"
31 +#endif
32 +
33 void a52_bit_allocate (a52_state_t * state, ba_t * ba, int bndstart,
34 int start, int end, int fastleak, int slowleak,
35 expbap_t * expbap);
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
55 @@ -31,6 +35,10 @@
57 #define BUFFER_SIZE 4096
59 +#ifdef ALT_BITSTREAM_READER
60 +int indx=0;
61 +#endif
62 +
63 void a52_bitstream_set_ptr (a52_state_t * state, uint8_t * buf)
64 {
65 int align;
66 @@ -38,6 +46,9 @@
67 align = (long)buf & 3;
68 state->buffer_start = (uint32_t *) (buf - align);
69 state->bits_left = 0;
70 +#ifdef ALT_BITSTREAM_READER
71 + indx=0;
72 +#endif
73 bitstream_get (state, align * 8);
74 }
76 --- liba52/bitstream.h 2006-06-12 15:05:07.000000000 +0200
77 +++ liba52/bitstream.h 2006-06-05 02:23:02.000000000 +0200
78 @@ -21,6 +25,42 @@
79 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
80 */
82 +/* code from ffmpeg/libavcodec */
83 +#if defined(__sparc__) || defined(hpux)
84 +/*
85 + * the alt bitstream reader performs unaligned memory accesses; that doesn't work
86 + * on sparc/hpux. For now, disable ALT_BITSTREAM_READER.
87 + */
88 +#undef ALT_BITSTREAM_READER
89 +#else
90 +// alternative (faster) bitstram reader (reades upto 3 bytes over the end of the input)
91 +#define ALT_BITSTREAM_READER
92 +
93 +/* used to avoid misaligned exceptions on some archs (alpha, ...) */
94 +#if ARCH_X86 || HAVE_ARMV6
95 +# define unaligned32(a) (*(uint32_t*)(a))
96 +#else
97 +# ifdef __GNUC__
98 +static inline uint32_t unaligned32(const void *v) {
99 + struct Unaligned {
100 + uint32_t i;
101 + } __attribute__((packed));
102 +
103 + return ((const struct Unaligned *) v)->i;
104 +}
105 +# elif defined(__DECC)
106 +static inline uint32_t unaligned32(const void *v) {
107 + return *(const __unaligned uint32_t *) v;
108 +}
109 +# else
110 +static inline uint32_t unaligned32(const void *v) {
111 + return *(const uint32_t *) v;
112 +}
113 +# endif
114 +#endif //!ARCH_X86
115 +
116 +#endif
117 +
118 /* (stolen from the kernel) */
119 #ifdef WORDS_BIGENDIAN
121 @@ -28,7 +68,7 @@
123 #else
125 -# if 0 && defined (__i386__)
126 +# if defined (__i386__)
128 # define swab32(x) __i386_swab32(x)
129 static inline const uint32_t __i386_swab32(uint32_t x)
130 @@ -39,19 +79,34 @@
132 # else
134 -# define swab32(x)\
135 -((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) | \
136 - (((uint8_t*)&x)[2] << 8) | (((uint8_t*)&x)[3]))
137 -
138 +# define swab32(x) __generic_swab32(x)
139 + static inline const uint32_t __generic_swab32(uint32_t x)
140 + {
141 + return ((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) |
142 + (((uint8_t*)&x)[2] << 8) | (((uint8_t*)&x)[3]));
143 + }
144 # endif
145 #endif
147 +#ifdef ALT_BITSTREAM_READER
148 +extern int indx;
149 +#endif
150 +
151 void a52_bitstream_set_ptr (a52_state_t * state, uint8_t * buf);
152 uint32_t a52_bitstream_get_bh (a52_state_t * state, uint32_t num_bits);
153 int32_t a52_bitstream_get_bh_2 (a52_state_t * state, uint32_t num_bits);
155 static inline uint32_t bitstream_get (a52_state_t * state, uint32_t num_bits)
156 {
157 +#ifdef ALT_BITSTREAM_READER
158 + uint32_t result= swab32( unaligned32(((uint8_t *)state->buffer_start)+(indx>>3)) );
159 +
160 + result<<= (indx&0x07);
161 + result>>= 32 - num_bits;
162 + indx+= num_bits;
163 +
164 + return result;
165 +#else
166 uint32_t result;
168 if (num_bits < state->bits_left) {
169 @@ -61,10 +116,29 @@
170 }
172 return a52_bitstream_get_bh (state, num_bits);
173 +#endif
174 +}
175 +
176 +static inline void bitstream_skip(a52_state_t * state, int num_bits)
177 +{
178 +#ifdef ALT_BITSTREAM_READER
179 + indx+= num_bits;
180 +#else
181 + bitstream_get(state, num_bits);
182 +#endif
183 }
185 static inline int32_t bitstream_get_2 (a52_state_t * state, uint32_t num_bits)
186 {
187 +#ifdef ALT_BITSTREAM_READER
188 + int32_t result= swab32( unaligned32(((uint8_t *)state->buffer_start)+(indx>>3)) );
189 +
190 + result<<= (indx&0x07);
191 + result>>= 32 - num_bits;
192 + indx+= num_bits;
193 +
194 + return result;
195 +#else
196 int32_t result;
198 if (num_bits < state->bits_left) {
199 @@ -74,4 +148,5 @@
200 }
202 return a52_bitstream_get_bh_2 (state, num_bits);
203 +#endif
204 }
205 --- liba52/downmix.c 2006-06-12 15:17:53.000000000 +0200
206 +++ liba52/downmix.c 2006-06-05 02:23:02.000000000 +0200
207 @@ -19,18 +23,46 @@
208 * You should have received a copy of the GNU General Public License
209 * along with this program; if not, write to the Free Software
210 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
211 + *
212 + * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
213 */
215 #include "config.h"
217 #include <string.h>
218 #include <inttypes.h>
220 #include "a52.h"
221 #include "a52_internal.h"
222 +#include "mm_accel.h"
224 #define CONVERT(acmod,output) (((output) << 3) + (acmod))
226 +
227 +void (*a52_downmix)(sample_t * samples, int acmod, int output, sample_t bias,
228 + sample_t clev, sample_t slev)= NULL;
229 +void (*a52_upmix)(sample_t * samples, int acmod, int output)= NULL;
230 +
231 +static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
232 + sample_t clev, sample_t slev);
233 +static void downmix_3dnow (sample_t * samples, int acmod, int output, sample_t bias,
234 + sample_t clev, sample_t slev);
235 +static void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
236 + sample_t clev, sample_t slev);
237 +static void upmix_MMX (sample_t * samples, int acmod, int output);
238 +static void upmix_C (sample_t * samples, int acmod, int output);
239 +
240 +void downmix_accel_init(uint32_t mm_accel)
241 +{
242 + a52_upmix= upmix_C;
243 + a52_downmix= downmix_C;
244 +#if ARCH_X86 || ARCH_X86_64
245 + if(mm_accel & MM_ACCEL_X86_MMX) a52_upmix= upmix_MMX;
246 + if(mm_accel & MM_ACCEL_X86_SSE) a52_downmix= downmix_SSE;
247 + if(mm_accel & MM_ACCEL_X86_3DNOW) a52_downmix= downmix_3dnow;
248 +#endif
249 +}
250 +
251 int a52_downmix_init (int input, int flags, sample_t * level,
252 sample_t clev, sample_t slev)
253 {
254 @@ -447,7 +479,7 @@
255 samples[i] = 0;
256 }
258 -void a52_downmix (sample_t * samples, int acmod, int output, sample_t bias,
259 +void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
260 sample_t clev, sample_t slev)
261 {
262 switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
263 @@ -559,7 +591,7 @@
264 break;
266 case CONVERT (A52_3F2R, A52_2F1R):
267 - mix3to2 (samples, bias);
268 + mix3to2 (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
269 move2to1 (samples + 768, samples + 512, bias);
270 break;
272 @@ -583,12 +615,12 @@
273 break;
275 case CONVERT (A52_3F1R, A52_3F2R):
276 - memcpy (samples + 1027, samples + 768, 256 * sizeof (sample_t));
277 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
278 break;
279 }
280 }
282 -void a52_upmix (sample_t * samples, int acmod, int output)
283 +void upmix_C (sample_t * samples, int acmod, int output)
284 {
285 switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
287 @@ -653,3 +685,1104 @@
288 goto mix_31to21;
289 }
290 }
291 +
292 +#if ARCH_X86 || ARCH_X86_64
293 +static void mix2to1_SSE (sample_t * dest, sample_t * src, sample_t bias)
294 +{
295 + __asm__ volatile(
296 + "movlps %2, %%xmm7 \n\t"
297 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
298 + "mov $-1024, %%"REG_S" \n\t"
299 + ASMALIGN(4)
300 + "1: \n\t"
301 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
302 + "movaps 16(%0, %%"REG_S"), %%xmm1\n\t"
303 + "addps (%1, %%"REG_S"), %%xmm0 \n\t"
304 + "addps 16(%1, %%"REG_S"), %%xmm1\n\t"
305 + "addps %%xmm7, %%xmm0 \n\t"
306 + "addps %%xmm7, %%xmm1 \n\t"
307 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
308 + "movaps %%xmm1, 16(%1, %%"REG_S")\n\t"
309 + "add $32, %%"REG_S" \n\t"
310 + " jnz 1b \n\t"
311 + :: "r" (src+256), "r" (dest+256), "m" (bias)
312 + : "%"REG_S
313 + );
314 +}
315 +
316 +static void mix3to1_SSE (sample_t * samples, sample_t bias)
317 +{
318 + __asm__ volatile(
319 + "movlps %1, %%xmm7 \n\t"
320 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
321 + "mov $-1024, %%"REG_S" \n\t"
322 + ASMALIGN(4)
323 + "1: \n\t"
324 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
325 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
326 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
327 + "addps %%xmm7, %%xmm1 \n\t"
328 + "addps %%xmm1, %%xmm0 \n\t"
329 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
330 + "add $16, %%"REG_S" \n\t"
331 + " jnz 1b \n\t"
332 + :: "r" (samples+256), "m" (bias)
333 + : "%"REG_S
334 + );
335 +}
336 +
337 +static void mix4to1_SSE (sample_t * samples, sample_t bias)
338 +{
339 + __asm__ volatile(
340 + "movlps %1, %%xmm7 \n\t"
341 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
342 + "mov $-1024, %%"REG_S" \n\t"
343 + ASMALIGN(4)
344 + "1: \n\t"
345 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
346 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
347 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
348 + "addps 3072(%0, %%"REG_S"), %%xmm1\n\t"
349 + "addps %%xmm7, %%xmm0 \n\t"
350 + "addps %%xmm1, %%xmm0 \n\t"
351 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
352 + "add $16, %%"REG_S" \n\t"
353 + " jnz 1b \n\t"
354 + :: "r" (samples+256), "m" (bias)
355 + : "%"REG_S
356 + );
357 +}
358 +
359 +static void mix5to1_SSE (sample_t * samples, sample_t bias)
360 +{
361 + __asm__ volatile(
362 + "movlps %1, %%xmm7 \n\t"
363 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
364 + "mov $-1024, %%"REG_S" \n\t"
365 + ASMALIGN(4)
366 + "1: \n\t"
367 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
368 + "movaps 1024(%0, %%"REG_S"), %%xmm1\n\t"
369 + "addps 2048(%0, %%"REG_S"), %%xmm0\n\t"
370 + "addps 3072(%0, %%"REG_S"), %%xmm1\n\t"
371 + "addps %%xmm7, %%xmm0 \n\t"
372 + "addps 4096(%0, %%"REG_S"), %%xmm1\n\t"
373 + "addps %%xmm1, %%xmm0 \n\t"
374 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
375 + "add $16, %%"REG_S" \n\t"
376 + " jnz 1b \n\t"
377 + :: "r" (samples+256), "m" (bias)
378 + : "%"REG_S
379 + );
380 +}
381 +
382 +static void mix3to2_SSE (sample_t * samples, sample_t bias)
383 +{
384 + __asm__ volatile(
385 + "movlps %1, %%xmm7 \n\t"
386 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
387 + "mov $-1024, %%"REG_S" \n\t"
388 + ASMALIGN(4)
389 + "1: \n\t"
390 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
391 + "addps %%xmm7, %%xmm0 \n\t" //common
392 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
393 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
394 + "addps %%xmm0, %%xmm1 \n\t"
395 + "addps %%xmm0, %%xmm2 \n\t"
396 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
397 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
398 + "add $16, %%"REG_S" \n\t"
399 + " jnz 1b \n\t"
400 + :: "r" (samples+256), "m" (bias)
401 + : "%"REG_S
402 + );
403 +}
404 +
405 +static void mix21to2_SSE (sample_t * left, sample_t * right, sample_t bias)
406 +{
407 + __asm__ volatile(
408 + "movlps %2, %%xmm7 \n\t"
409 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
410 + "mov $-1024, %%"REG_S" \n\t"
411 + ASMALIGN(4)
412 + "1: \n\t"
413 + "movaps 1024(%1, %%"REG_S"), %%xmm0\n\t"
414 + "addps %%xmm7, %%xmm0 \n\t" //common
415 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
416 + "movaps (%1, %%"REG_S"), %%xmm2 \n\t"
417 + "addps %%xmm0, %%xmm1 \n\t"
418 + "addps %%xmm0, %%xmm2 \n\t"
419 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
420 + "movaps %%xmm2, (%1, %%"REG_S") \n\t"
421 + "add $16, %%"REG_S" \n\t"
422 + " jnz 1b \n\t"
423 + :: "r" (left+256), "r" (right+256), "m" (bias)
424 + : "%"REG_S
425 + );
426 +}
427 +
428 +static void mix21toS_SSE (sample_t * samples, sample_t bias)
429 +{
430 + __asm__ volatile(
431 + "movlps %1, %%xmm7 \n\t"
432 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
433 + "mov $-1024, %%"REG_S" \n\t"
434 + ASMALIGN(4)
435 + "1: \n\t"
436 + "movaps 2048(%0, %%"REG_S"), %%xmm0\n\t" // surround
437 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
438 + "movaps 1024(%0, %%"REG_S"), %%xmm2\n\t"
439 + "addps %%xmm7, %%xmm1 \n\t"
440 + "addps %%xmm7, %%xmm2 \n\t"
441 + "subps %%xmm0, %%xmm1 \n\t"
442 + "addps %%xmm0, %%xmm2 \n\t"
443 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
444 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
445 + "add $16, %%"REG_S" \n\t"
446 + " jnz 1b \n\t"
447 + :: "r" (samples+256), "m" (bias)
448 + : "%"REG_S
449 + );
450 +}
451 +
452 +static void mix31to2_SSE (sample_t * samples, sample_t bias)
453 +{
454 + __asm__ volatile(
455 + "movlps %1, %%xmm7 \n\t"
456 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
457 + "mov $-1024, %%"REG_S" \n\t"
458 + ASMALIGN(4)
459 + "1: \n\t"
460 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
461 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t"
462 + "addps %%xmm7, %%xmm0 \n\t" // common
463 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
464 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
465 + "addps %%xmm0, %%xmm1 \n\t"
466 + "addps %%xmm0, %%xmm2 \n\t"
467 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
468 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
469 + "add $16, %%"REG_S" \n\t"
470 + " jnz 1b \n\t"
471 + :: "r" (samples+256), "m" (bias)
472 + : "%"REG_S
473 + );
474 +}
475 +
476 +static void mix31toS_SSE (sample_t * samples, sample_t bias)
477 +{
478 + __asm__ volatile(
479 + "movlps %1, %%xmm7 \n\t"
480 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
481 + "mov $-1024, %%"REG_S" \n\t"
482 + ASMALIGN(4)
483 + "1: \n\t"
484 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
485 + "movaps 3072(%0, %%"REG_S"), %%xmm3\n\t" // surround
486 + "addps %%xmm7, %%xmm0 \n\t" // common
487 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
488 + "movaps 2048(%0, %%"REG_S"), %%xmm2\n\t"
489 + "addps %%xmm0, %%xmm1 \n\t"
490 + "addps %%xmm0, %%xmm2 \n\t"
491 + "subps %%xmm3, %%xmm1 \n\t"
492 + "addps %%xmm3, %%xmm2 \n\t"
493 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
494 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
495 + "add $16, %%"REG_S" \n\t"
496 + " jnz 1b \n\t"
497 + :: "r" (samples+256), "m" (bias)
498 + : "%"REG_S
499 + );
500 +}
501 +
502 +static void mix22toS_SSE (sample_t * samples, sample_t bias)
503 +{
504 + __asm__ volatile(
505 + "movlps %1, %%xmm7 \n\t"
506 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
507 + "mov $-1024, %%"REG_S" \n\t"
508 + ASMALIGN(4)
509 + "1: \n\t"
510 + "movaps 2048(%0, %%"REG_S"), %%xmm0\n\t"
511 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t" // surround
512 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
513 + "movaps 1024(%0, %%"REG_S"), %%xmm2\n\t"
514 + "addps %%xmm7, %%xmm1 \n\t"
515 + "addps %%xmm7, %%xmm2 \n\t"
516 + "subps %%xmm0, %%xmm1 \n\t"
517 + "addps %%xmm0, %%xmm2 \n\t"
518 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
519 + "movaps %%xmm2, 1024(%0, %%"REG_S")\n\t"
520 + "add $16, %%"REG_S" \n\t"
521 + " jnz 1b \n\t"
522 + :: "r" (samples+256), "m" (bias)
523 + : "%"REG_S
524 + );
525 +}
526 +
527 +static void mix32to2_SSE (sample_t * samples, sample_t bias)
528 +{
529 + __asm__ volatile(
530 + "movlps %1, %%xmm7 \n\t"
531 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
532 + "mov $-1024, %%"REG_S" \n\t"
533 + ASMALIGN(4)
534 + "1: \n\t"
535 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
536 + "addps %%xmm7, %%xmm0 \n\t" // common
537 + "movaps %%xmm0, %%xmm1 \n\t" // common
538 + "addps (%0, %%"REG_S"), %%xmm0 \n\t"
539 + "addps 2048(%0, %%"REG_S"), %%xmm1\n\t"
540 + "addps 3072(%0, %%"REG_S"), %%xmm0\n\t"
541 + "addps 4096(%0, %%"REG_S"), %%xmm1\n\t"
542 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
543 + "movaps %%xmm1, 1024(%0, %%"REG_S")\n\t"
544 + "add $16, %%"REG_S" \n\t"
545 + " jnz 1b \n\t"
546 + :: "r" (samples+256), "m" (bias)
547 + : "%"REG_S
548 + );
549 +}
550 +
551 +static void mix32toS_SSE (sample_t * samples, sample_t bias)
552 +{
553 + __asm__ volatile(
554 + "movlps %1, %%xmm7 \n\t"
555 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
556 + "mov $-1024, %%"REG_S" \n\t"
557 + ASMALIGN(4)
558 + "1: \n\t"
559 + "movaps 1024(%0, %%"REG_S"), %%xmm0\n\t"
560 + "movaps 3072(%0, %%"REG_S"), %%xmm2\n\t"
561 + "addps %%xmm7, %%xmm0 \n\t" // common
562 + "addps 4096(%0, %%"REG_S"), %%xmm2\n\t" // surround
563 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
564 + "movaps 2048(%0, %%"REG_S"), %%xmm3\n\t"
565 + "subps %%xmm2, %%xmm1 \n\t"
566 + "addps %%xmm2, %%xmm3 \n\t"
567 + "addps %%xmm0, %%xmm1 \n\t"
568 + "addps %%xmm0, %%xmm3 \n\t"
569 + "movaps %%xmm1, (%0, %%"REG_S") \n\t"
570 + "movaps %%xmm3, 1024(%0, %%"REG_S")\n\t"
571 + "add $16, %%"REG_S" \n\t"
572 + " jnz 1b \n\t"
573 + :: "r" (samples+256), "m" (bias)
574 + : "%"REG_S
575 + );
576 +}
577 +
578 +static void move2to1_SSE (sample_t * src, sample_t * dest, sample_t bias)
579 +{
580 + __asm__ volatile(
581 + "movlps %2, %%xmm7 \n\t"
582 + "shufps $0x00, %%xmm7, %%xmm7 \n\t"
583 + "mov $-1024, %%"REG_S" \n\t"
584 + ASMALIGN(4)
585 + "1: \n\t"
586 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
587 + "movaps 16(%0, %%"REG_S"), %%xmm1\n\t"
588 + "addps 1024(%0, %%"REG_S"), %%xmm0\n\t"
589 + "addps 1040(%0, %%"REG_S"), %%xmm1\n\t"
590 + "addps %%xmm7, %%xmm0 \n\t"
591 + "addps %%xmm7, %%xmm1 \n\t"
592 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
593 + "movaps %%xmm1, 16(%1, %%"REG_S")\n\t"
594 + "add $32, %%"REG_S" \n\t"
595 + " jnz 1b \n\t"
596 + :: "r" (src+256), "r" (dest+256), "m" (bias)
597 + : "%"REG_S
598 + );
599 +}
600 +
601 +static void zero_MMX(sample_t * samples)
602 +{
603 + __asm__ volatile(
604 + "mov $-1024, %%"REG_S" \n\t"
605 + "pxor %%mm0, %%mm0 \n\t"
606 + ASMALIGN(4)
607 + "1: \n\t"
608 + "movq %%mm0, (%0, %%"REG_S") \n\t"
609 + "movq %%mm0, 8(%0, %%"REG_S") \n\t"
610 + "movq %%mm0, 16(%0, %%"REG_S") \n\t"
611 + "movq %%mm0, 24(%0, %%"REG_S") \n\t"
612 + "add $32, %%"REG_S" \n\t"
613 + " jnz 1b \n\t"
614 + "emms"
615 + :: "r" (samples+256)
616 + : "%"REG_S
617 + );
618 +}
619 +
620 +static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
621 + sample_t clev, sample_t slev)
622 +{
623 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
624 +
625 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
626 + memcpy (samples, samples + 256, 256 * sizeof (sample_t));
627 + break;
628 +
629 + case CONVERT (A52_CHANNEL, A52_MONO):
630 + case CONVERT (A52_STEREO, A52_MONO):
631 + mix_2to1_SSE:
632 + mix2to1_SSE (samples, samples + 256, bias);
633 + break;
634 +
635 + case CONVERT (A52_2F1R, A52_MONO):
636 + if (slev == 0)
637 + goto mix_2to1_SSE;
638 + case CONVERT (A52_3F, A52_MONO):
639 + mix_3to1_SSE:
640 + mix3to1_SSE (samples, bias);
641 + break;
642 +
643 + case CONVERT (A52_3F1R, A52_MONO):
644 + if (slev == 0)
645 + goto mix_3to1_SSE;
646 + case CONVERT (A52_2F2R, A52_MONO):
647 + if (slev == 0)
648 + goto mix_2to1_SSE;
649 + mix4to1_SSE (samples, bias);
650 + break;
651 +
652 + case CONVERT (A52_3F2R, A52_MONO):
653 + if (slev == 0)
654 + goto mix_3to1_SSE;
655 + mix5to1_SSE (samples, bias);
656 + break;
657 +
658 + case CONVERT (A52_MONO, A52_DOLBY):
659 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
660 + break;
661 +
662 + case CONVERT (A52_3F, A52_STEREO):
663 + case CONVERT (A52_3F, A52_DOLBY):
664 + mix_3to2_SSE:
665 + mix3to2_SSE (samples, bias);
666 + break;
667 +
668 + case CONVERT (A52_2F1R, A52_STEREO):
669 + if (slev == 0)
670 + break;
671 + mix21to2_SSE (samples, samples + 256, bias);
672 + break;
673 +
674 + case CONVERT (A52_2F1R, A52_DOLBY):
675 + mix21toS_SSE (samples, bias);
676 + break;
677 +
678 + case CONVERT (A52_3F1R, A52_STEREO):
679 + if (slev == 0)
680 + goto mix_3to2_SSE;
681 + mix31to2_SSE (samples, bias);
682 + break;
683 +
684 + case CONVERT (A52_3F1R, A52_DOLBY):
685 + mix31toS_SSE (samples, bias);
686 + break;
687 +
688 + case CONVERT (A52_2F2R, A52_STEREO):
689 + if (slev == 0)
690 + break;
691 + mix2to1_SSE (samples, samples + 512, bias);
692 + mix2to1_SSE (samples + 256, samples + 768, bias);
693 + break;
694 +
695 + case CONVERT (A52_2F2R, A52_DOLBY):
696 + mix22toS_SSE (samples, bias);
697 + break;
698 +
699 + case CONVERT (A52_3F2R, A52_STEREO):
700 + if (slev == 0)
701 + goto mix_3to2_SSE;
702 + mix32to2_SSE (samples, bias);
703 + break;
704 +
705 + case CONVERT (A52_3F2R, A52_DOLBY):
706 + mix32toS_SSE (samples, bias);
707 + break;
708 +
709 + case CONVERT (A52_3F1R, A52_3F):
710 + if (slev == 0)
711 + break;
712 + mix21to2_SSE (samples, samples + 512, bias);
713 + break;
714 +
715 + case CONVERT (A52_3F2R, A52_3F):
716 + if (slev == 0)
717 + break;
718 + mix2to1_SSE (samples, samples + 768, bias);
719 + mix2to1_SSE (samples + 512, samples + 1024, bias);
720 + break;
721 +
722 + case CONVERT (A52_3F1R, A52_2F1R):
723 + mix3to2_SSE (samples, bias);
724 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
725 + break;
726 +
727 + case CONVERT (A52_2F2R, A52_2F1R):
728 + mix2to1_SSE (samples + 512, samples + 768, bias);
729 + break;
730 +
731 + case CONVERT (A52_3F2R, A52_2F1R):
732 + mix3to2_SSE (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
733 + move2to1_SSE (samples + 768, samples + 512, bias);
734 + break;
735 +
736 + case CONVERT (A52_3F2R, A52_3F1R):
737 + mix2to1_SSE (samples + 768, samples + 1024, bias);
738 + break;
739 +
740 + case CONVERT (A52_2F1R, A52_2F2R):
741 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
742 + break;
743 +
744 + case CONVERT (A52_3F1R, A52_2F2R):
745 + mix3to2_SSE (samples, bias);
746 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
747 + break;
748 +
749 + case CONVERT (A52_3F2R, A52_2F2R):
750 + mix3to2_SSE (samples, bias);
751 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
752 + memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
753 + break;
754 +
755 + case CONVERT (A52_3F1R, A52_3F2R):
756 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
757 + break;
758 + }
759 +}
760 +
761 +static void upmix_MMX (sample_t * samples, int acmod, int output)
762 +{
763 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
764 +
765 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
766 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
767 + break;
768 +
769 + case CONVERT (A52_3F2R, A52_MONO):
770 + zero_MMX (samples + 1024);
771 + case CONVERT (A52_3F1R, A52_MONO):
772 + case CONVERT (A52_2F2R, A52_MONO):
773 + zero_MMX (samples + 768);
774 + case CONVERT (A52_3F, A52_MONO):
775 + case CONVERT (A52_2F1R, A52_MONO):
776 + zero_MMX (samples + 512);
777 + case CONVERT (A52_CHANNEL, A52_MONO):
778 + case CONVERT (A52_STEREO, A52_MONO):
779 + zero_MMX (samples + 256);
780 + break;
781 +
782 + case CONVERT (A52_3F2R, A52_STEREO):
783 + case CONVERT (A52_3F2R, A52_DOLBY):
784 + zero_MMX (samples + 1024);
785 + case CONVERT (A52_3F1R, A52_STEREO):
786 + case CONVERT (A52_3F1R, A52_DOLBY):
787 + zero_MMX (samples + 768);
788 + case CONVERT (A52_3F, A52_STEREO):
789 + case CONVERT (A52_3F, A52_DOLBY):
790 + mix_3to2_MMX:
791 + memcpy (samples + 512, samples + 256, 256 * sizeof (sample_t));
792 + zero_MMX (samples + 256);
793 + break;
794 +
795 + case CONVERT (A52_2F2R, A52_STEREO):
796 + case CONVERT (A52_2F2R, A52_DOLBY):
797 + zero_MMX (samples + 768);
798 + case CONVERT (A52_2F1R, A52_STEREO):
799 + case CONVERT (A52_2F1R, A52_DOLBY):
800 + zero_MMX (samples + 512);
801 + break;
802 +
803 + case CONVERT (A52_3F2R, A52_3F):
804 + zero_MMX (samples + 1024);
805 + case CONVERT (A52_3F1R, A52_3F):
806 + case CONVERT (A52_2F2R, A52_2F1R):
807 + zero_MMX (samples + 768);
808 + break;
809 +
810 + case CONVERT (A52_3F2R, A52_3F1R):
811 + zero_MMX (samples + 1024);
812 + break;
813 +
814 + case CONVERT (A52_3F2R, A52_2F1R):
815 + zero_MMX (samples + 1024);
816 + case CONVERT (A52_3F1R, A52_2F1R):
817 + mix_31to21_MMX:
818 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
819 + goto mix_3to2_MMX;
820 +
821 + case CONVERT (A52_3F2R, A52_2F2R):
822 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
823 + goto mix_31to21_MMX;
824 + }
825 +}
826 +
827 +static void mix2to1_3dnow (sample_t * dest, sample_t * src, sample_t bias)
828 +{
829 + __asm__ volatile(
830 + "movd %2, %%mm7 \n\t"
831 + "punpckldq %2, %%mm7 \n\t"
832 + "mov $-1024, %%"REG_S" \n\t"
833 + ASMALIGN(4)
834 + "1: \n\t"
835 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
836 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
837 + "movq 16(%0, %%"REG_S"), %%mm2 \n\t"
838 + "movq 24(%0, %%"REG_S"), %%mm3 \n\t"
839 + "pfadd (%1, %%"REG_S"), %%mm0 \n\t"
840 + "pfadd 8(%1, %%"REG_S"), %%mm1 \n\t"
841 + "pfadd 16(%1, %%"REG_S"), %%mm2 \n\t"
842 + "pfadd 24(%1, %%"REG_S"), %%mm3 \n\t"
843 + "pfadd %%mm7, %%mm0 \n\t"
844 + "pfadd %%mm7, %%mm1 \n\t"
845 + "pfadd %%mm7, %%mm2 \n\t"
846 + "pfadd %%mm7, %%mm3 \n\t"
847 + "movq %%mm0, (%1, %%"REG_S") \n\t"
848 + "movq %%mm1, 8(%1, %%"REG_S") \n\t"
849 + "movq %%mm2, 16(%1, %%"REG_S") \n\t"
850 + "movq %%mm3, 24(%1, %%"REG_S") \n\t"
851 + "add $32, %%"REG_S" \n\t"
852 + " jnz 1b \n\t"
853 + :: "r" (src+256), "r" (dest+256), "m" (bias)
854 + : "%"REG_S
855 + );
856 +}
857 +
858 +static void mix3to1_3dnow (sample_t * samples, sample_t bias)
859 +{
860 + __asm__ volatile(
861 + "movd %1, %%mm7 \n\t"
862 + "punpckldq %1, %%mm7 \n\t"
863 + "mov $-1024, %%"REG_S" \n\t"
864 + ASMALIGN(4)
865 + "1: \n\t"
866 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
867 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
868 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
869 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
870 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
871 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
872 + "pfadd %%mm7, %%mm0 \n\t"
873 + "pfadd %%mm7, %%mm1 \n\t"
874 + "pfadd %%mm2, %%mm0 \n\t"
875 + "pfadd %%mm3, %%mm1 \n\t"
876 + "movq %%mm0, (%0, %%"REG_S") \n\t"
877 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
878 + "add $16, %%"REG_S" \n\t"
879 + " jnz 1b \n\t"
880 + :: "r" (samples+256), "m" (bias)
881 + : "%"REG_S
882 + );
883 +}
884 +
885 +static void mix4to1_3dnow (sample_t * samples, sample_t bias)
886 +{
887 + __asm__ volatile(
888 + "movd %1, %%mm7 \n\t"
889 + "punpckldq %1, %%mm7 \n\t"
890 + "mov $-1024, %%"REG_S" \n\t"
891 + ASMALIGN(4)
892 + "1: \n\t"
893 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
894 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
895 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
896 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
897 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
898 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
899 + "pfadd 3072(%0, %%"REG_S"), %%mm2\n\t"
900 + "pfadd 3080(%0, %%"REG_S"), %%mm3\n\t"
901 + "pfadd %%mm7, %%mm0 \n\t"
902 + "pfadd %%mm7, %%mm1 \n\t"
903 + "pfadd %%mm2, %%mm0 \n\t"
904 + "pfadd %%mm3, %%mm1 \n\t"
905 + "movq %%mm0, (%0, %%"REG_S") \n\t"
906 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
907 + "add $16, %%"REG_S" \n\t"
908 + " jnz 1b \n\t"
909 + :: "r" (samples+256), "m" (bias)
910 + : "%"REG_S
911 + );
912 +}
913 +
914 +static void mix5to1_3dnow (sample_t * samples, sample_t bias)
915 +{
916 + __asm__ volatile(
917 + "movd %1, %%mm7 \n\t"
918 + "punpckldq %1, %%mm7 \n\t"
919 + "mov $-1024, %%"REG_S" \n\t"
920 + ASMALIGN(4)
921 + "1: \n\t"
922 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
923 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
924 + "movq 1024(%0, %%"REG_S"), %%mm2\n\t"
925 + "movq 1032(%0, %%"REG_S"), %%mm3\n\t"
926 + "pfadd 2048(%0, %%"REG_S"), %%mm0\n\t"
927 + "pfadd 2056(%0, %%"REG_S"), %%mm1\n\t"
928 + "pfadd 3072(%0, %%"REG_S"), %%mm2\n\t"
929 + "pfadd 3080(%0, %%"REG_S"), %%mm3\n\t"
930 + "pfadd %%mm7, %%mm0 \n\t"
931 + "pfadd %%mm7, %%mm1 \n\t"
932 + "pfadd 4096(%0, %%"REG_S"), %%mm2\n\t"
933 + "pfadd 4104(%0, %%"REG_S"), %%mm3\n\t"
934 + "pfadd %%mm2, %%mm0 \n\t"
935 + "pfadd %%mm3, %%mm1 \n\t"
936 + "movq %%mm0, (%0, %%"REG_S") \n\t"
937 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
938 + "add $16, %%"REG_S" \n\t"
939 + " jnz 1b \n\t"
940 + :: "r" (samples+256), "m" (bias)
941 + : "%"REG_S
942 + );
943 +}
944 +
945 +static void mix3to2_3dnow (sample_t * samples, sample_t bias)
946 +{
947 + __asm__ volatile(
948 + "movd %1, %%mm7 \n\t"
949 + "punpckldq %1, %%mm7 \n\t"
950 + "mov $-1024, %%"REG_S" \n\t"
951 + ASMALIGN(4)
952 + "1: \n\t"
953 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
954 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
955 + "pfadd %%mm7, %%mm0 \n\t" //common
956 + "pfadd %%mm7, %%mm1 \n\t" //common
957 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
958 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
959 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
960 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
961 + "pfadd %%mm0, %%mm2 \n\t"
962 + "pfadd %%mm1, %%mm3 \n\t"
963 + "pfadd %%mm0, %%mm4 \n\t"
964 + "pfadd %%mm1, %%mm5 \n\t"
965 + "movq %%mm2, (%0, %%"REG_S") \n\t"
966 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
967 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
968 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
969 + "add $16, %%"REG_S" \n\t"
970 + " jnz 1b \n\t"
971 + :: "r" (samples+256), "m" (bias)
972 + : "%"REG_S
973 + );
974 +}
975 +
976 +static void mix21to2_3dnow (sample_t * left, sample_t * right, sample_t bias)
977 +{
978 + __asm__ volatile(
979 + "movd %2, %%mm7 \n\t"
980 + "punpckldq %2, %%mm7 \n\t"
981 + "mov $-1024, %%"REG_S" \n\t"
982 + ASMALIGN(4)
983 + "1: \n\t"
984 + "movq 1024(%1, %%"REG_S"), %%mm0\n\t"
985 + "movq 1032(%1, %%"REG_S"), %%mm1\n\t"
986 + "pfadd %%mm7, %%mm0 \n\t" //common
987 + "pfadd %%mm7, %%mm1 \n\t" //common
988 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
989 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
990 + "movq (%1, %%"REG_S"), %%mm4 \n\t"
991 + "movq 8(%1, %%"REG_S"), %%mm5 \n\t"
992 + "pfadd %%mm0, %%mm2 \n\t"
993 + "pfadd %%mm1, %%mm3 \n\t"
994 + "pfadd %%mm0, %%mm4 \n\t"
995 + "pfadd %%mm1, %%mm5 \n\t"
996 + "movq %%mm2, (%0, %%"REG_S") \n\t"
997 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
998 + "movq %%mm4, (%1, %%"REG_S") \n\t"
999 + "movq %%mm5, 8(%1, %%"REG_S") \n\t"
1000 + "add $16, %%"REG_S" \n\t"
1001 + " jnz 1b \n\t"
1002 + :: "r" (left+256), "r" (right+256), "m" (bias)
1003 + : "%"REG_S
1004 + );
1005 +}
1006 +
1007 +static void mix21toS_3dnow (sample_t * samples, sample_t bias)
1008 +{
1009 + __asm__ volatile(
1010 + "movd %1, %%mm7 \n\t"
1011 + "punpckldq %1, %%mm7 \n\t"
1012 + "mov $-1024, %%"REG_S" \n\t"
1013 + ASMALIGN(4)
1014 + "1: \n\t"
1015 + "movq 2048(%0, %%"REG_S"), %%mm0\n\t" // surround
1016 + "movq 2056(%0, %%"REG_S"), %%mm1\n\t" // surround
1017 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1018 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1019 + "movq 1024(%0, %%"REG_S"), %%mm4\n\t"
1020 + "movq 1032(%0, %%"REG_S"), %%mm5\n\t"
1021 + "pfadd %%mm7, %%mm2 \n\t"
1022 + "pfadd %%mm7, %%mm3 \n\t"
1023 + "pfadd %%mm7, %%mm4 \n\t"
1024 + "pfadd %%mm7, %%mm5 \n\t"
1025 + "pfsub %%mm0, %%mm2 \n\t"
1026 + "pfsub %%mm1, %%mm3 \n\t"
1027 + "pfadd %%mm0, %%mm4 \n\t"
1028 + "pfadd %%mm1, %%mm5 \n\t"
1029 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1030 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1031 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1032 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1033 + "add $16, %%"REG_S" \n\t"
1034 + " jnz 1b \n\t"
1035 + :: "r" (samples+256), "m" (bias)
1036 + : "%"REG_S
1037 + );
1038 +}
1039 +
1040 +static void mix31to2_3dnow (sample_t * samples, sample_t bias)
1041 +{
1042 + __asm__ volatile(
1043 + "movd %1, %%mm7 \n\t"
1044 + "punpckldq %1, %%mm7 \n\t"
1045 + "mov $-1024, %%"REG_S" \n\t"
1046 + ASMALIGN(4)
1047 + "1: \n\t"
1048 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1049 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1050 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t"
1051 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t"
1052 + "pfadd %%mm7, %%mm0 \n\t" // common
1053 + "pfadd %%mm7, %%mm1 \n\t" // common
1054 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1055 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1056 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
1057 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
1058 + "pfadd %%mm0, %%mm2 \n\t"
1059 + "pfadd %%mm1, %%mm3 \n\t"
1060 + "pfadd %%mm0, %%mm4 \n\t"
1061 + "pfadd %%mm1, %%mm5 \n\t"
1062 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1063 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1064 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1065 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1066 + "add $16, %%"REG_S" \n\t"
1067 + " jnz 1b \n\t"
1068 + :: "r" (samples+256), "m" (bias)
1069 + : "%"REG_S
1070 + );
1071 +}
1072 +
1073 +static void mix31toS_3dnow (sample_t * samples, sample_t bias)
1074 +{
1075 + __asm__ volatile(
1076 + "movd %1, %%mm7 \n\t"
1077 + "punpckldq %1, %%mm7 \n\t"
1078 + "mov $-1024, %%"REG_S" \n\t"
1079 + ASMALIGN(4)
1080 + "1: \n\t"
1081 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1082 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1083 + "pfadd %%mm7, %%mm0 \n\t" // common
1084 + "pfadd %%mm7, %%mm1 \n\t" // common
1085 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1086 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1087 + "movq 2048(%0, %%"REG_S"), %%mm4\n\t"
1088 + "movq 2056(%0, %%"REG_S"), %%mm5\n\t"
1089 + "pfadd %%mm0, %%mm2 \n\t"
1090 + "pfadd %%mm1, %%mm3 \n\t"
1091 + "pfadd %%mm0, %%mm4 \n\t"
1092 + "pfadd %%mm1, %%mm5 \n\t"
1093 + "movq 3072(%0, %%"REG_S"), %%mm0\n\t" // surround
1094 + "movq 3080(%0, %%"REG_S"), %%mm1\n\t" // surround
1095 + "pfsub %%mm0, %%mm2 \n\t"
1096 + "pfsub %%mm1, %%mm3 \n\t"
1097 + "pfadd %%mm0, %%mm4 \n\t"
1098 + "pfadd %%mm1, %%mm5 \n\t"
1099 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1100 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1101 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1102 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1103 + "add $16, %%"REG_S" \n\t"
1104 + " jnz 1b \n\t"
1105 + :: "r" (samples+256), "m" (bias)
1106 + : "%"REG_S
1107 + );
1108 +}
1109 +
1110 +static void mix22toS_3dnow (sample_t * samples, sample_t bias)
1111 +{
1112 + __asm__ volatile(
1113 + "movd %1, %%mm7 \n\t"
1114 + "punpckldq %1, %%mm7 \n\t"
1115 + "mov $-1024, %%"REG_S" \n\t"
1116 + ASMALIGN(4)
1117 + "1: \n\t"
1118 + "movq 2048(%0, %%"REG_S"), %%mm0\n\t"
1119 + "movq 2056(%0, %%"REG_S"), %%mm1\n\t"
1120 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t" // surround
1121 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t" // surround
1122 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1123 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1124 + "movq 1024(%0, %%"REG_S"), %%mm4\n\t"
1125 + "movq 1032(%0, %%"REG_S"), %%mm5\n\t"
1126 + "pfadd %%mm7, %%mm2 \n\t"
1127 + "pfadd %%mm7, %%mm3 \n\t"
1128 + "pfadd %%mm7, %%mm4 \n\t"
1129 + "pfadd %%mm7, %%mm5 \n\t"
1130 + "pfsub %%mm0, %%mm2 \n\t"
1131 + "pfsub %%mm1, %%mm3 \n\t"
1132 + "pfadd %%mm0, %%mm4 \n\t"
1133 + "pfadd %%mm1, %%mm5 \n\t"
1134 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1135 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1136 + "movq %%mm4, 1024(%0, %%"REG_S")\n\t"
1137 + "movq %%mm5, 1032(%0, %%"REG_S")\n\t"
1138 + "add $16, %%"REG_S" \n\t"
1139 + " jnz 1b \n\t"
1140 + :: "r" (samples+256), "m" (bias)
1141 + : "%"REG_S
1142 + );
1143 +}
1144 +
1145 +static void mix32to2_3dnow (sample_t * samples, sample_t bias)
1146 +{
1147 + __asm__ volatile(
1148 + "movd %1, %%mm7 \n\t"
1149 + "punpckldq %1, %%mm7 \n\t"
1150 + "mov $-1024, %%"REG_S" \n\t"
1151 + ASMALIGN(4)
1152 + "1: \n\t"
1153 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1154 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1155 + "pfadd %%mm7, %%mm0 \n\t" // common
1156 + "pfadd %%mm7, %%mm1 \n\t" // common
1157 + "movq %%mm0, %%mm2 \n\t" // common
1158 + "movq %%mm1, %%mm3 \n\t" // common
1159 + "pfadd (%0, %%"REG_S"), %%mm0 \n\t"
1160 + "pfadd 8(%0, %%"REG_S"), %%mm1 \n\t"
1161 + "pfadd 2048(%0, %%"REG_S"), %%mm2\n\t"
1162 + "pfadd 2056(%0, %%"REG_S"), %%mm3\n\t"
1163 + "pfadd 3072(%0, %%"REG_S"), %%mm0\n\t"
1164 + "pfadd 3080(%0, %%"REG_S"), %%mm1\n\t"
1165 + "pfadd 4096(%0, %%"REG_S"), %%mm2\n\t"
1166 + "pfadd 4104(%0, %%"REG_S"), %%mm3\n\t"
1167 + "movq %%mm0, (%0, %%"REG_S") \n\t"
1168 + "movq %%mm1, 8(%0, %%"REG_S") \n\t"
1169 + "movq %%mm2, 1024(%0, %%"REG_S")\n\t"
1170 + "movq %%mm3, 1032(%0, %%"REG_S")\n\t"
1171 + "add $16, %%"REG_S" \n\t"
1172 + " jnz 1b \n\t"
1173 + :: "r" (samples+256), "m" (bias)
1174 + : "%"REG_S
1175 + );
1176 +}
1177 +
1178 +/* todo: should be optimized better */
1179 +static void mix32toS_3dnow (sample_t * samples, sample_t bias)
1180 +{
1181 + __asm__ volatile(
1182 + "mov $-1024, %%"REG_S" \n\t"
1183 + ASMALIGN(4)
1184 + "1: \n\t"
1185 + "movd %1, %%mm7 \n\t"
1186 + "punpckldq %1, %%mm7 \n\t"
1187 + "movq 1024(%0, %%"REG_S"), %%mm0\n\t"
1188 + "movq 1032(%0, %%"REG_S"), %%mm1\n\t"
1189 + "movq 3072(%0, %%"REG_S"), %%mm4\n\t"
1190 + "movq 3080(%0, %%"REG_S"), %%mm5\n\t"
1191 + "pfadd %%mm7, %%mm0 \n\t" // common
1192 + "pfadd %%mm7, %%mm1 \n\t" // common
1193 + "pfadd 4096(%0, %%"REG_S"), %%mm4\n\t" // surround
1194 + "pfadd 4104(%0, %%"REG_S"), %%mm5\n\t" // surround
1195 + "movq (%0, %%"REG_S"), %%mm2 \n\t"
1196 + "movq 8(%0, %%"REG_S"), %%mm3 \n\t"
1197 + "movq 2048(%0, %%"REG_S"), %%mm6\n\t"
1198 + "movq 2056(%0, %%"REG_S"), %%mm7\n\t"
1199 + "pfsub %%mm4, %%mm2 \n\t"
1200 + "pfsub %%mm5, %%mm3 \n\t"
1201 + "pfadd %%mm4, %%mm6 \n\t"
1202 + "pfadd %%mm5, %%mm7 \n\t"
1203 + "pfadd %%mm0, %%mm2 \n\t"
1204 + "pfadd %%mm1, %%mm3 \n\t"
1205 + "pfadd %%mm0, %%mm6 \n\t"
1206 + "pfadd %%mm1, %%mm7 \n\t"
1207 + "movq %%mm2, (%0, %%"REG_S") \n\t"
1208 + "movq %%mm3, 8(%0, %%"REG_S") \n\t"
1209 + "movq %%mm6, 1024(%0, %%"REG_S")\n\t"
1210 + "movq %%mm7, 1032(%0, %%"REG_S")\n\t"
1211 + "add $16, %%"REG_S" \n\t"
1212 + " jnz 1b \n\t"
1213 + :: "r" (samples+256), "m" (bias)
1214 + : "%"REG_S
1215 + );
1216 +}
1217 +
1218 +static void move2to1_3dnow (sample_t * src, sample_t * dest, sample_t bias)
1219 +{
1220 + __asm__ volatile(
1221 + "movd %2, %%mm7 \n\t"
1222 + "punpckldq %2, %%mm7 \n\t"
1223 + "mov $-1024, %%"REG_S" \n\t"
1224 + ASMALIGN(4)
1225 + "1: \n\t"
1226 + "movq (%0, %%"REG_S"), %%mm0 \n\t"
1227 + "movq 8(%0, %%"REG_S"), %%mm1 \n\t"
1228 + "movq 16(%0, %%"REG_S"), %%mm2 \n\t"
1229 + "movq 24(%0, %%"REG_S"), %%mm3 \n\t"
1230 + "pfadd 1024(%0, %%"REG_S"), %%mm0\n\t"
1231 + "pfadd 1032(%0, %%"REG_S"), %%mm1\n\t"
1232 + "pfadd 1040(%0, %%"REG_S"), %%mm2\n\t"
1233 + "pfadd 1048(%0, %%"REG_S"), %%mm3\n\t"
1234 + "pfadd %%mm7, %%mm0 \n\t"
1235 + "pfadd %%mm7, %%mm1 \n\t"
1236 + "pfadd %%mm7, %%mm2 \n\t"
1237 + "pfadd %%mm7, %%mm3 \n\t"
1238 + "movq %%mm0, (%1, %%"REG_S") \n\t"
1239 + "movq %%mm1, 8(%1, %%"REG_S") \n\t"
1240 + "movq %%mm2, 16(%1, %%"REG_S") \n\t"
1241 + "movq %%mm3, 24(%1, %%"REG_S") \n\t"
1242 + "add $32, %%"REG_S" \n\t"
1243 + " jnz 1b \n\t"
1244 + :: "r" (src+256), "r" (dest+256), "m" (bias)
1245 + : "%"REG_S
1246 + );
1247 +}
1248 +
1249 +static void downmix_3dnow (sample_t * samples, int acmod, int output, sample_t bias,
1250 + sample_t clev, sample_t slev)
1251 +{
1252 + switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
1253 +
1254 + case CONVERT (A52_CHANNEL, A52_CHANNEL2):
1255 + memcpy (samples, samples + 256, 256 * sizeof (sample_t));
1256 + break;
1257 +
1258 + case CONVERT (A52_CHANNEL, A52_MONO):
1259 + case CONVERT (A52_STEREO, A52_MONO):
1260 + mix_2to1_3dnow:
1261 + mix2to1_3dnow (samples, samples + 256, bias);
1262 + break;
1263 +
1264 + case CONVERT (A52_2F1R, A52_MONO):
1265 + if (slev == 0)
1266 + goto mix_2to1_3dnow;
1267 + case CONVERT (A52_3F, A52_MONO):
1268 + mix_3to1_3dnow:
1269 + mix3to1_3dnow (samples, bias);
1270 + break;
1271 +
1272 + case CONVERT (A52_3F1R, A52_MONO):
1273 + if (slev == 0)
1274 + goto mix_3to1_3dnow;
1275 + case CONVERT (A52_2F2R, A52_MONO):
1276 + if (slev == 0)
1277 + goto mix_2to1_3dnow;
1278 + mix4to1_3dnow (samples, bias);
1279 + break;
1280 +
1281 + case CONVERT (A52_3F2R, A52_MONO):
1282 + if (slev == 0)
1283 + goto mix_3to1_3dnow;
1284 + mix5to1_3dnow (samples, bias);
1285 + break;
1286 +
1287 + case CONVERT (A52_MONO, A52_DOLBY):
1288 + memcpy (samples + 256, samples, 256 * sizeof (sample_t));
1289 + break;
1290 +
1291 + case CONVERT (A52_3F, A52_STEREO):
1292 + case CONVERT (A52_3F, A52_DOLBY):
1293 + mix_3to2_3dnow:
1294 + mix3to2_3dnow (samples, bias);
1295 + break;
1296 +
1297 + case CONVERT (A52_2F1R, A52_STEREO):
1298 + if (slev == 0)
1299 + break;
1300 + mix21to2_3dnow (samples, samples + 256, bias);
1301 + break;
1302 +
1303 + case CONVERT (A52_2F1R, A52_DOLBY):
1304 + mix21toS_3dnow (samples, bias);
1305 + break;
1306 +
1307 + case CONVERT (A52_3F1R, A52_STEREO):
1308 + if (slev == 0)
1309 + goto mix_3to2_3dnow;
1310 + mix31to2_3dnow (samples, bias);
1311 + break;
1312 +
1313 + case CONVERT (A52_3F1R, A52_DOLBY):
1314 + mix31toS_3dnow (samples, bias);
1315 + break;
1316 +
1317 + case CONVERT (A52_2F2R, A52_STEREO):
1318 + if (slev == 0)
1319 + break;
1320 + mix2to1_3dnow (samples, samples + 512, bias);
1321 + mix2to1_3dnow (samples + 256, samples + 768, bias);
1322 + break;
1323 +
1324 + case CONVERT (A52_2F2R, A52_DOLBY):
1325 + mix22toS_3dnow (samples, bias);
1326 + break;
1327 +
1328 + case CONVERT (A52_3F2R, A52_STEREO):
1329 + if (slev == 0)
1330 + goto mix_3to2_3dnow;
1331 + mix32to2_3dnow (samples, bias);
1332 + break;
1333 +
1334 + case CONVERT (A52_3F2R, A52_DOLBY):
1335 + mix32toS_3dnow (samples, bias);
1336 + break;
1337 +
1338 + case CONVERT (A52_3F1R, A52_3F):
1339 + if (slev == 0)
1340 + break;
1341 + mix21to2_3dnow (samples, samples + 512, bias);
1342 + break;
1343 +
1344 + case CONVERT (A52_3F2R, A52_3F):
1345 + if (slev == 0)
1346 + break;
1347 + mix2to1_3dnow (samples, samples + 768, bias);
1348 + mix2to1_3dnow (samples + 512, samples + 1024, bias);
1349 + break;
1350 +
1351 + case CONVERT (A52_3F1R, A52_2F1R):
1352 + mix3to2_3dnow (samples, bias);
1353 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1354 + break;
1355 +
1356 + case CONVERT (A52_2F2R, A52_2F1R):
1357 + mix2to1_3dnow (samples + 512, samples + 768, bias);
1358 + break;
1359 +
1360 + case CONVERT (A52_3F2R, A52_2F1R):
1361 + mix3to2_3dnow (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
1362 + move2to1_3dnow (samples + 768, samples + 512, bias);
1363 + break;
1364 +
1365 + case CONVERT (A52_3F2R, A52_3F1R):
1366 + mix2to1_3dnow (samples + 768, samples + 1024, bias);
1367 + break;
1368 +
1369 + case CONVERT (A52_2F1R, A52_2F2R):
1370 + memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
1371 + break;
1372 +
1373 + case CONVERT (A52_3F1R, A52_2F2R):
1374 + mix3to2_3dnow (samples, bias);
1375 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1376 + break;
1377 +
1378 + case CONVERT (A52_3F2R, A52_2F2R):
1379 + mix3to2_3dnow (samples, bias);
1380 + memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
1381 + memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
1382 + break;
1383 +
1384 + case CONVERT (A52_3F1R, A52_3F2R):
1385 + memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
1386 + break;
1387 + }
1388 + __asm__ volatile("femms":::"memory");
1389 +}
1390 +
1391 +#endif // ARCH_X86 || ARCH_X86_64
1392 --- liba52/imdct.c 2008-02-19 00:18:33.000000000 +0100
1393 +++ liba52/imdct.c 2008-02-19 00:16:40.000000000 +0100
1394 @@ -22,6 +26,11 @@
1395 * You should have received a copy of the GNU General Public License
1396 * along with this program; if not, write to the Free Software
1397 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
1398 + *
1399 + * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
1400 + * 3DNOW optimizations from Nick Kurshev <nickols_k@mail.ru>
1401 + * michael did port them from libac3 (untested, perhaps totally broken)
1402 + * AltiVec optimizations from Romain Dolbeau (romain@dolbeau.org)
1403 */
1405 #include "config.h"
1406 @@ -39,12 +48,50 @@
1407 #include "a52.h"
1408 #include "a52_internal.h"
1409 #include "mm_accel.h"
1410 +#include "mangle.h"
1411 +
1412 +void (*a52_imdct_512) (sample_t * data, sample_t * delay, sample_t bias);
1413 +
1414 +#if CONFIG_RUNTIME_CPUDETECT
1415 +#undef HAVE_AMD3DNOWEXT
1416 +#define HAVE_AMD3DNOWEXT 0
1417 +#endif
1419 typedef struct complex_s {
1420 sample_t real;
1421 sample_t imag;
1422 } complex_t;
1424 +static const int pm128[128] attribute_used __attribute__((aligned(16))) =
1425 +{
1426 + 0, 16, 32, 48, 64, 80, 96, 112, 8, 40, 72, 104, 24, 56, 88, 120,
1427 + 4, 20, 36, 52, 68, 84, 100, 116, 12, 28, 44, 60, 76, 92, 108, 124,
1428 + 2, 18, 34, 50, 66, 82, 98, 114, 10, 42, 74, 106, 26, 58, 90, 122,
1429 + 6, 22, 38, 54, 70, 86, 102, 118, 14, 46, 78, 110, 30, 62, 94, 126,
1430 + 1, 17, 33, 49, 65, 81, 97, 113, 9, 41, 73, 105, 25, 57, 89, 121,
1431 + 5, 21, 37, 53, 69, 85, 101, 117, 13, 29, 45, 61, 77, 93, 109, 125,
1432 + 3, 19, 35, 51, 67, 83, 99, 115, 11, 43, 75, 107, 27, 59, 91, 123,
1433 + 7, 23, 39, 55, 71, 87, 103, 119, 15, 31, 47, 63, 79, 95, 111, 127
1434 +};
1435 +
1436 +static uint8_t attribute_used bit_reverse_512[] = {
1437 + 0x00, 0x40, 0x20, 0x60, 0x10, 0x50, 0x30, 0x70,
1438 + 0x08, 0x48, 0x28, 0x68, 0x18, 0x58, 0x38, 0x78,
1439 + 0x04, 0x44, 0x24, 0x64, 0x14, 0x54, 0x34, 0x74,
1440 + 0x0c, 0x4c, 0x2c, 0x6c, 0x1c, 0x5c, 0x3c, 0x7c,
1441 + 0x02, 0x42, 0x22, 0x62, 0x12, 0x52, 0x32, 0x72,
1442 + 0x0a, 0x4a, 0x2a, 0x6a, 0x1a, 0x5a, 0x3a, 0x7a,
1443 + 0x06, 0x46, 0x26, 0x66, 0x16, 0x56, 0x36, 0x76,
1444 + 0x0e, 0x4e, 0x2e, 0x6e, 0x1e, 0x5e, 0x3e, 0x7e,
1445 + 0x01, 0x41, 0x21, 0x61, 0x11, 0x51, 0x31, 0x71,
1446 + 0x09, 0x49, 0x29, 0x69, 0x19, 0x59, 0x39, 0x79,
1447 + 0x05, 0x45, 0x25, 0x65, 0x15, 0x55, 0x35, 0x75,
1448 + 0x0d, 0x4d, 0x2d, 0x6d, 0x1d, 0x5d, 0x3d, 0x7d,
1449 + 0x03, 0x43, 0x23, 0x63, 0x13, 0x53, 0x33, 0x73,
1450 + 0x0b, 0x4b, 0x2b, 0x6b, 0x1b, 0x5b, 0x3b, 0x7b,
1451 + 0x07, 0x47, 0x27, 0x67, 0x17, 0x57, 0x37, 0x77,
1452 + 0x0f, 0x4f, 0x2f, 0x6f, 0x1f, 0x5f, 0x3f, 0x7f};
1453 +
1454 static uint8_t fftorder[] = {
1455 0,128, 64,192, 32,160,224, 96, 16,144, 80,208,240,112, 48,176,
1456 8,136, 72,200, 40,168,232,104,248,120, 56,184, 24,152,216, 88,
1457 @@ -56,6 +103,40 @@
1458 6,134, 70,198, 38,166,230,102,246,118, 54,182, 22,150,214, 86
1459 };
1461 +static complex_t __attribute__((aligned(16))) buf[128];
1462 +
1463 +/* Twiddle factor LUT */
1464 +static complex_t __attribute__((aligned(16))) w_1[1];
1465 +static complex_t __attribute__((aligned(16))) w_2[2];
1466 +static complex_t __attribute__((aligned(16))) w_4[4];
1467 +static complex_t __attribute__((aligned(16))) w_8[8];
1468 +static complex_t __attribute__((aligned(16))) w_16[16];
1469 +static complex_t __attribute__((aligned(16))) w_32[32];
1470 +static complex_t __attribute__((aligned(16))) w_64[64];
1471 +static complex_t __attribute__((aligned(16))) * w[7] = {w_1, w_2, w_4, w_8, w_16, w_32, w_64};
1472 +
1473 +/* Twiddle factors for IMDCT */
1474 +static sample_t __attribute__((aligned(16))) xcos1[128];
1475 +static sample_t __attribute__((aligned(16))) xsin1[128];
1476 +
1477 +#if ARCH_X86 || ARCH_X86_64
1478 +// NOTE: SSE needs 16byte alignment or it will segfault
1479 +//
1480 +static float __attribute__((aligned(16))) sseSinCos1c[256];
1481 +static float __attribute__((aligned(16))) sseSinCos1d[256];
1482 +static float attribute_used __attribute__((aligned(16))) ps111_1[4]={1,1,1,-1};
1483 +//static float __attribute__((aligned(16))) sseW0[4];
1484 +static float __attribute__((aligned(16))) sseW1[8];
1485 +static float __attribute__((aligned(16))) sseW2[16];
1486 +static float __attribute__((aligned(16))) sseW3[32];
1487 +static float __attribute__((aligned(16))) sseW4[64];
1488 +static float __attribute__((aligned(16))) sseW5[128];
1489 +static float __attribute__((aligned(16))) sseW6[256];
1490 +static float __attribute__((aligned(16))) *sseW[7]=
1491 + {NULL /*sseW0*/,sseW1,sseW2,sseW3,sseW4,sseW5,sseW6};
1492 +static float __attribute__((aligned(16))) sseWindow[512];
1493 +#endif
1494 +
1495 /* Root values for IFFT */
1496 static sample_t roots16[3];
1497 static sample_t roots32[7];
1498 @@ -241,7 +322,7 @@
1499 ifft_pass (buf, roots128 - 32, 32);
1500 }
1502 -void a52_imdct_512 (sample_t * data, sample_t * delay, sample_t bias)
1503 +void imdct_do_512 (sample_t * data, sample_t * delay, sample_t bias)
1504 {
1505 int i, k;
1506 sample_t t_r, t_i, a_r, a_i, b_r, b_i, w_1, w_2;
1507 @@ -285,6 +366,702 @@
1508 }
1509 }
1511 +#if HAVE_ALTIVEC
1512 +
1513 +#ifdef HAVE_ALTIVEC_H
1514 +#include <altivec.h>
1515 +#endif
1516 +
1517 +// used to build registers permutation vectors (vcprm)
1518 +// the 's' are for words in the _s_econd vector
1519 +#define WORD_0 0x00,0x01,0x02,0x03
1520 +#define WORD_1 0x04,0x05,0x06,0x07
1521 +#define WORD_2 0x08,0x09,0x0a,0x0b
1522 +#define WORD_3 0x0c,0x0d,0x0e,0x0f
1523 +#define WORD_s0 0x10,0x11,0x12,0x13
1524 +#define WORD_s1 0x14,0x15,0x16,0x17
1525 +#define WORD_s2 0x18,0x19,0x1a,0x1b
1526 +#define WORD_s3 0x1c,0x1d,0x1e,0x1f
1527 +
1528 +#define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
1529 +#define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
1530 +
1531 +#define FOUROF(a) {a,a,a,a}
1532 +
1533 +// vcprmle is used to keep the same index as in the SSE version.
1534 +// it's the same as vcprm, with the index inversed
1535 +// ('le' is Little Endian)
1536 +#define vcprmle(a,b,c,d) vcprm(d,c,b,a)
1537 +
1538 +// used to build inverse/identity vectors (vcii)
1539 +// n is _n_egative, p is _p_ositive
1540 +#define FLOAT_n -1.
1541 +#define FLOAT_p 1.
1542 +
1543 +
1544 +void
1545 +imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
1546 +{
1547 + int i;
1548 + int k;
1549 + int p,q;
1550 + int m;
1551 + long two_m;
1552 + long two_m_plus_one;
1553 +
1554 + sample_t tmp_b_i;
1555 + sample_t tmp_b_r;
1556 + sample_t tmp_a_i;
1557 + sample_t tmp_a_r;
1558 +
1559 + sample_t *data_ptr;
1560 + sample_t *delay_ptr;
1561 + sample_t *window_ptr;
1562 +
1563 + /* 512 IMDCT with source and dest data in 'data' */
1564 +
1565 + /* Pre IFFT complex multiply plus IFFT cmplx conjugate & reordering*/
1566 + for( i=0; i < 128; i++) {
1567 + /* z[i] = (X[256-2*i-1] + j * X[2*i]) * (xcos1[i] + j * xsin1[i]) ; */
1568 + int j= bit_reverse_512[i];
1569 + buf[i].real = (data[256-2*j-1] * xcos1[j]) - (data[2*j] * xsin1[j]);
1570 + buf[i].imag = -1.0 * ((data[2*j] * xcos1[j]) + (data[256-2*j-1] * xsin1[j]));
1571 + }
1572 +
1573 + /* 1. iteration */
1574 + for(i = 0; i < 128; i += 2) {
1575 +#if 0
1576 + tmp_a_r = buf[i].real;
1577 + tmp_a_i = buf[i].imag;
1578 + tmp_b_r = buf[i+1].real;
1579 + tmp_b_i = buf[i+1].imag;
1580 + buf[i].real = tmp_a_r + tmp_b_r;
1581 + buf[i].imag = tmp_a_i + tmp_b_i;
1582 + buf[i+1].real = tmp_a_r - tmp_b_r;
1583 + buf[i+1].imag = tmp_a_i - tmp_b_i;
1584 +#else
1585 + vector float temp, bufv;
1586 +
1587 + bufv = vec_ld(i << 3, (float*)buf);
1588 + temp = vec_perm(bufv, bufv, vcprm(2,3,0,1));
1589 + bufv = vec_madd(bufv, vcii(p,p,n,n), temp);
1590 + vec_st(bufv, i << 3, (float*)buf);
1591 +#endif
1592 + }
1593 +
1594 + /* 2. iteration */
1595 + // Note w[1]={{1,0}, {0,-1}}
1596 + for(i = 0; i < 128; i += 4) {
1597 +#if 0
1598 + tmp_a_r = buf[i].real;
1599 + tmp_a_i = buf[i].imag;
1600 + tmp_b_r = buf[i+2].real;
1601 + tmp_b_i = buf[i+2].imag;
1602 + buf[i].real = tmp_a_r + tmp_b_r;
1603 + buf[i].imag = tmp_a_i + tmp_b_i;
1604 + buf[i+2].real = tmp_a_r - tmp_b_r;
1605 + buf[i+2].imag = tmp_a_i - tmp_b_i;
1606 + tmp_a_r = buf[i+1].real;
1607 + tmp_a_i = buf[i+1].imag;
1608 + /* WARNING: im <-> re here ! */
1609 + tmp_b_r = buf[i+3].imag;
1610 + tmp_b_i = buf[i+3].real;
1611 + buf[i+1].real = tmp_a_r + tmp_b_r;
1612 + buf[i+1].imag = tmp_a_i - tmp_b_i;
1613 + buf[i+3].real = tmp_a_r - tmp_b_r;
1614 + buf[i+3].imag = tmp_a_i + tmp_b_i;
1615 +#else
1616 + vector float buf01, buf23, temp1, temp2;
1617 +
1618 + buf01 = vec_ld((i + 0) << 3, (float*)buf);
1619 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1620 + buf23 = vec_perm(buf23,buf23,vcprm(0,1,3,2));
1621 +
1622 + temp1 = vec_madd(buf23, vcii(p,p,p,n), buf01);
1623 + temp2 = vec_madd(buf23, vcii(n,n,n,p), buf01);
1624 +
1625 + vec_st(temp1, (i + 0) << 3, (float*)buf);
1626 + vec_st(temp2, (i + 2) << 3, (float*)buf);
1627 +#endif
1628 + }
1629 +
1630 + /* 3. iteration */
1631 + for(i = 0; i < 128; i += 8) {
1632 +#if 0
1633 + tmp_a_r = buf[i].real;
1634 + tmp_a_i = buf[i].imag;
1635 + tmp_b_r = buf[i+4].real;
1636 + tmp_b_i = buf[i+4].imag;
1637 + buf[i].real = tmp_a_r + tmp_b_r;
1638 + buf[i].imag = tmp_a_i + tmp_b_i;
1639 + buf[i+4].real = tmp_a_r - tmp_b_r;
1640 + buf[i+4].imag = tmp_a_i - tmp_b_i;
1641 + tmp_a_r = buf[1+i].real;
1642 + tmp_a_i = buf[1+i].imag;
1643 + tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
1644 + tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
1645 + buf[1+i].real = tmp_a_r + tmp_b_r;
1646 + buf[1+i].imag = tmp_a_i + tmp_b_i;
1647 + buf[i+5].real = tmp_a_r - tmp_b_r;
1648 + buf[i+5].imag = tmp_a_i - tmp_b_i;
1649 + tmp_a_r = buf[i+2].real;
1650 + tmp_a_i = buf[i+2].imag;
1651 + /* WARNING re <-> im & sign */
1652 + tmp_b_r = buf[i+6].imag;
1653 + tmp_b_i = - buf[i+6].real;
1654 + buf[i+2].real = tmp_a_r + tmp_b_r;
1655 + buf[i+2].imag = tmp_a_i + tmp_b_i;
1656 + buf[i+6].real = tmp_a_r - tmp_b_r;
1657 + buf[i+6].imag = tmp_a_i - tmp_b_i;
1658 + tmp_a_r = buf[i+3].real;
1659 + tmp_a_i = buf[i+3].imag;
1660 + tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
1661 + tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
1662 + buf[i+3].real = tmp_a_r + tmp_b_r;
1663 + buf[i+3].imag = tmp_a_i + tmp_b_i;
1664 + buf[i+7].real = tmp_a_r - tmp_b_r;
1665 + buf[i+7].imag = tmp_a_i - tmp_b_i;
1666 +#else
1667 + vector float buf01, buf23, buf45, buf67;
1668 +
1669 + buf01 = vec_ld((i + 0) << 3, (float*)buf);
1670 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1671 +
1672 + tmp_b_r = (buf[i+5].real + buf[i+5].imag) * w[2][1].real;
1673 + tmp_b_i = (buf[i+5].imag - buf[i+5].real) * w[2][1].real;
1674 + buf[i+5].real = tmp_b_r;
1675 + buf[i+5].imag = tmp_b_i;
1676 + tmp_b_r = (buf[i+7].real - buf[i+7].imag) * w[2][3].imag;
1677 + tmp_b_i = (buf[i+7].imag + buf[i+7].real) * w[2][3].imag;
1678 + buf[i+7].real = tmp_b_r;
1679 + buf[i+7].imag = tmp_b_i;
1680 +
1681 + buf23 = vec_ld((i + 2) << 3, (float*)buf);
1682 + buf45 = vec_ld((i + 4) << 3, (float*)buf);
1683 + buf67 = vec_ld((i + 6) << 3, (float*)buf);
1684 + buf67 = vec_perm(buf67, buf67, vcprm(1,0,2,3));
1685 +
1686 + vec_st(vec_add(buf01, buf45), (i + 0) << 3, (float*)buf);
1687 + vec_st(vec_madd(buf67, vcii(p,n,p,p), buf23), (i + 2) << 3, (float*)buf);
1688 + vec_st(vec_sub(buf01, buf45), (i + 4) << 3, (float*)buf);
1689 + vec_st(vec_nmsub(buf67, vcii(p,n,p,p), buf23), (i + 6) << 3, (float*)buf);
1690 +#endif
1691 + }
1692 +
1693 + /* 4-7. iterations */
1694 + for (m=3; m < 7; m++) {
1695 + two_m = (1 << m);
1696 +
1697 + two_m_plus_one = two_m<<1;
1698 +
1699 + for(i = 0; i < 128; i += two_m_plus_one) {
1700 + for(k = 0; k < two_m; k+=2) {
1701 +#if 0
1702 + int p = k + i;
1703 + int q = p + two_m;
1704 + tmp_a_r = buf[p].real;
1705 + tmp_a_i = buf[p].imag;
1706 + tmp_b_r =
1707 + buf[q].real * w[m][k].real -
1708 + buf[q].imag * w[m][k].imag;
1709 + tmp_b_i =
1710 + buf[q].imag * w[m][k].real +
1711 + buf[q].real * w[m][k].imag;
1712 + buf[p].real = tmp_a_r + tmp_b_r;
1713 + buf[p].imag = tmp_a_i + tmp_b_i;
1714 + buf[q].real = tmp_a_r - tmp_b_r;
1715 + buf[q].imag = tmp_a_i - tmp_b_i;
1716 +
1717 + tmp_a_r = buf[(p + 1)].real;
1718 + tmp_a_i = buf[(p + 1)].imag;
1719 + tmp_b_r =
1720 + buf[(q + 1)].real * w[m][(k + 1)].real -
1721 + buf[(q + 1)].imag * w[m][(k + 1)].imag;
1722 + tmp_b_i =
1723 + buf[(q + 1)].imag * w[m][(k + 1)].real +
1724 + buf[(q + 1)].real * w[m][(k + 1)].imag;
1725 + buf[(p + 1)].real = tmp_a_r + tmp_b_r;
1726 + buf[(p + 1)].imag = tmp_a_i + tmp_b_i;
1727 + buf[(q + 1)].real = tmp_a_r - tmp_b_r;
1728 + buf[(q + 1)].imag = tmp_a_i - tmp_b_i;
1729 +#else
1730 + int p = k + i;
1731 + int q = p + two_m;
1732 + vector float vecp, vecq, vecw, temp1, temp2, temp3, temp4;
1733 + const vector float vczero = (const vector float)FOUROF(0.);
1734 + // first compute buf[q] and buf[q+1]
1735 + vecq = vec_ld(q << 3, (float*)buf);
1736 + vecw = vec_ld(0, (float*)&(w[m][k]));
1737 + temp1 = vec_madd(vecq, vecw, vczero);
1738 + temp2 = vec_perm(vecq, vecq, vcprm(1,0,3,2));
1739 + temp2 = vec_madd(temp2, vecw, vczero);
1740 + temp3 = vec_perm(temp1, temp2, vcprm(0,s0,2,s2));
1741 + temp4 = vec_perm(temp1, temp2, vcprm(1,s1,3,s3));
1742 + vecq = vec_madd(temp4, vcii(n,p,n,p), temp3);
1743 + // then butterfly with buf[p] and buf[p+1]
1744 + vecp = vec_ld(p << 3, (float*)buf);
1745 +
1746 + temp1 = vec_add(vecp, vecq);
1747 + temp2 = vec_sub(vecp, vecq);
1748 +
1749 + vec_st(temp1, p << 3, (float*)buf);
1750 + vec_st(temp2, q << 3, (float*)buf);
1751 +#endif
1752 + }
1753 + }
1754 + }
1755 +
1756 + /* Post IFFT complex multiply plus IFFT complex conjugate*/
1757 + for( i=0; i < 128; i+=4) {
1758 + /* y[n] = z[n] * (xcos1[n] + j * xsin1[n]) ; */
1759 +#if 0
1760 + tmp_a_r = buf[(i + 0)].real;
1761 + tmp_a_i = -1.0 * buf[(i + 0)].imag;
1762 + buf[(i + 0)].real =
1763 + (tmp_a_r * xcos1[(i + 0)]) - (tmp_a_i * xsin1[(i + 0)]);
1764 + buf[(i + 0)].imag =
1765 + (tmp_a_r * xsin1[(i + 0)]) + (tmp_a_i * xcos1[(i + 0)]);
1766 +
1767 + tmp_a_r = buf[(i + 1)].real;
1768 + tmp_a_i = -1.0 * buf[(i + 1)].imag;
1769 + buf[(i + 1)].real =
1770 + (tmp_a_r * xcos1[(i + 1)]) - (tmp_a_i * xsin1[(i + 1)]);
1771 + buf[(i + 1)].imag =
1772 + (tmp_a_r * xsin1[(i + 1)]) + (tmp_a_i * xcos1[(i + 1)]);
1773 +
1774 + tmp_a_r = buf[(i + 2)].real;
1775 + tmp_a_i = -1.0 * buf[(i + 2)].imag;
1776 + buf[(i + 2)].real =
1777 + (tmp_a_r * xcos1[(i + 2)]) - (tmp_a_i * xsin1[(i + 2)]);
1778 + buf[(i + 2)].imag =
1779 + (tmp_a_r * xsin1[(i + 2)]) + (tmp_a_i * xcos1[(i + 2)]);
1780 +
1781 + tmp_a_r = buf[(i + 3)].real;
1782 + tmp_a_i = -1.0 * buf[(i + 3)].imag;
1783 + buf[(i + 3)].real =
1784 + (tmp_a_r * xcos1[(i + 3)]) - (tmp_a_i * xsin1[(i + 3)]);
1785 + buf[(i + 3)].imag =
1786 + (tmp_a_r * xsin1[(i + 3)]) + (tmp_a_i * xcos1[(i + 3)]);
1787 +#else
1788 + vector float bufv_0, bufv_2, cosv, sinv, temp1, temp2;
1789 + vector float temp0022, temp1133, tempCS01;
1790 + const vector float vczero = (const vector float)FOUROF(0.);
1791 +
1792 + bufv_0 = vec_ld((i + 0) << 3, (float*)buf);
1793 + bufv_2 = vec_ld((i + 2) << 3, (float*)buf);
1794 +
1795 + cosv = vec_ld(i << 2, xcos1);
1796 + sinv = vec_ld(i << 2, xsin1);
1797 +
1798 + temp0022 = vec_perm(bufv_0, bufv_0, vcprm(0,0,2,2));
1799 + temp1133 = vec_perm(bufv_0, bufv_0, vcprm(1,1,3,3));
1800 + tempCS01 = vec_perm(cosv, sinv, vcprm(0,s0,1,s1));
1801 + temp1 = vec_madd(temp0022, tempCS01, vczero);
1802 + tempCS01 = vec_perm(cosv, sinv, vcprm(s0,0,s1,1));
1803 + temp2 = vec_madd(temp1133, tempCS01, vczero);
1804 + bufv_0 = vec_madd(temp2, vcii(p,n,p,n), temp1);
1805 +
1806 + vec_st(bufv_0, (i + 0) << 3, (float*)buf);
1807 +
1808 + /* idem with bufv_2 and high-order cosv/sinv */
1809 +
1810 + temp0022 = vec_perm(bufv_2, bufv_2, vcprm(0,0,2,2));
1811 + temp1133 = vec_perm(bufv_2, bufv_2, vcprm(1,1,3,3));
1812 + tempCS01 = vec_perm(cosv, sinv, vcprm(2,s2,3,s3));
1813 + temp1 = vec_madd(temp0022, tempCS01, vczero);
1814 + tempCS01 = vec_perm(cosv, sinv, vcprm(s2,2,s3,3));
1815 + temp2 = vec_madd(temp1133, tempCS01, vczero);
1816 + bufv_2 = vec_madd(temp2, vcii(p,n,p,n), temp1);
1817 +
1818 + vec_st(bufv_2, (i + 2) << 3, (float*)buf);
1819 +
1820 +#endif
1821 + }
1822 +
1823 + data_ptr = data;
1824 + delay_ptr = delay;
1825 + window_ptr = a52_imdct_window;
1826 +
1827 + /* Window and convert to real valued signal */
1828 + for(i=0; i< 64; i++) {
1829 + *data_ptr++ = -buf[64+i].imag * *window_ptr++ + *delay_ptr++ + bias;
1830 + *data_ptr++ = buf[64-i-1].real * *window_ptr++ + *delay_ptr++ + bias;
1831 + }
1832 +
1833 + for(i=0; i< 64; i++) {
1834 + *data_ptr++ = -buf[i].real * *window_ptr++ + *delay_ptr++ + bias;
1835 + *data_ptr++ = buf[128-i-1].imag * *window_ptr++ + *delay_ptr++ + bias;
1836 + }
1837 +
1838 + /* The trailing edge of the window goes into the delay line */
1839 + delay_ptr = delay;
1840 +
1841 + for(i=0; i< 64; i++) {
1842 + *delay_ptr++ = -buf[64+i].real * *--window_ptr;
1843 + *delay_ptr++ = buf[64-i-1].imag * *--window_ptr;
1844 + }
1845 +
1846 + for(i=0; i<64; i++) {
1847 + *delay_ptr++ = buf[i].imag * *--window_ptr;
1848 + *delay_ptr++ = -buf[128-i-1].real * *--window_ptr;
1849 + }
1850 +}
1851 +#endif
1852 +
1853 +
1854 +// Stuff below this line is borrowed from libac3
1855 +#include "srfftp.h"
1856 +#if ARCH_X86 || ARCH_X86_64
1857 +#undef HAVE_AMD3DNOW
1858 +#define HAVE_AMD3DNOW 1
1859 +#include "srfftp_3dnow.h"
1860 +
1861 +const i_cmplx_t x_plus_minus_3dnow __attribute__ ((aligned (8))) = {{ 0x00000000UL, 0x80000000UL }};
1862 +const i_cmplx_t x_minus_plus_3dnow __attribute__ ((aligned (8))) = {{ 0x80000000UL, 0x00000000UL }};
1863 +const complex_t HSQRT2_3DNOW __attribute__ ((aligned (8))) = { 0.707106781188, 0.707106781188 };
1864 +
1865 +#undef HAVE_AMD3DNOWEXT
1866 +#define HAVE_AMD3DNOWEXT 0
1867 +#include "imdct_3dnow.h"
1868 +#undef HAVE_AMD3DNOWEXT
1869 +#define HAVE_AMD3DNOWEXT 1
1870 +#include "imdct_3dnow.h"
1871 +
1872 +void
1873 +imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
1874 +{
1875 +/* int i,k;
1876 + int p,q;*/
1877 + int m;
1878 + long two_m;
1879 + long two_m_plus_one;
1880 + long two_m_plus_one_shl3;
1881 + complex_t *buf_offset;
1882 +
1883 +/* sample_t tmp_a_i;
1884 + sample_t tmp_a_r;
1885 + sample_t tmp_b_i;
1886 + sample_t tmp_b_r;*/
1887 +
1888 + sample_t *data_ptr;
1889 + sample_t *delay_ptr;
1890 + sample_t *window_ptr;
1891 +
1892 + /* 512 IMDCT with source and dest data in 'data' */
1893 + /* see the c version (dct_do_512()), its allmost identical, just in C */
1894 +
1895 + /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
1896 + /* Bit reversed shuffling */
1897 + __asm__ volatile(
1898 + "xor %%"REG_S", %%"REG_S" \n\t"
1899 + "lea "MANGLE(bit_reverse_512)", %%"REG_a"\n\t"
1900 + "mov $1008, %%"REG_D" \n\t"
1901 + "push %%"REG_BP" \n\t" //use ebp without telling gcc
1902 + ASMALIGN(4)
1903 + "1: \n\t"
1904 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // XXXI
1905 + "movhps 8(%0, %%"REG_D"), %%xmm0 \n\t" // RXXI
1906 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // XXXi
1907 + "movhps (%0, %%"REG_D"), %%xmm1 \n\t" // rXXi
1908 + "shufps $0x33, %%xmm1, %%xmm0 \n\t" // irIR
1909 + "movaps "MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm2\n\t"
1910 + "mulps %%xmm0, %%xmm2 \n\t"
1911 + "shufps $0xB1, %%xmm0, %%xmm0 \n\t" // riRI
1912 + "mulps "MANGLE(sseSinCos1d)"(%%"REG_S"), %%xmm0\n\t"
1913 + "subps %%xmm0, %%xmm2 \n\t"
1914 + "movzb (%%"REG_a"), %%"REG_d" \n\t"
1915 + "movzb 1(%%"REG_a"), %%"REG_BP" \n\t"
1916 + "movlps %%xmm2, (%1, %%"REG_d", 8) \n\t"
1917 + "movhps %%xmm2, (%1, %%"REG_BP", 8) \n\t"
1918 + "add $16, %%"REG_S" \n\t"
1919 + "add $2, %%"REG_a" \n\t" // avoid complex addressing for P4 crap
1920 + "sub $16, %%"REG_D" \n\t"
1921 + "jnc 1b \n\t"
1922 + "pop %%"REG_BP" \n\t"//no we didnt touch ebp *g*
1923 + :: "b" (data), "c" (buf)
1924 + : "%"REG_S, "%"REG_D, "%"REG_a, "%"REG_d
1925 + );
1926 +
1927 +
1928 + /* FFT Merge */
1929 +/* unoptimized variant
1930 + for (m=1; m < 7; m++) {
1931 + if(m)
1932 + two_m = (1 << m);
1933 + else
1934 + two_m = 1;
1935 +
1936 + two_m_plus_one = (1 << (m+1));
1937 +
1938 + for(i = 0; i < 128; i += two_m_plus_one) {
1939 + for(k = 0; k < two_m; k++) {
1940 + p = k + i;
1941 + q = p + two_m;
1942 + tmp_a_r = buf[p].real;
1943 + tmp_a_i = buf[p].imag;
1944 + tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
1945 + tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
1946 + buf[p].real = tmp_a_r + tmp_b_r;
1947 + buf[p].imag = tmp_a_i + tmp_b_i;
1948 + buf[q].real = tmp_a_r - tmp_b_r;
1949 + buf[q].imag = tmp_a_i - tmp_b_i;
1950 + }
1951 + }
1952 + }
1953 +*/
1954 +
1955 + /* 1. iteration */
1956 + // Note w[0][0]={1,0}
1957 + __asm__ volatile(
1958 + "xorps %%xmm1, %%xmm1 \n\t"
1959 + "xorps %%xmm2, %%xmm2 \n\t"
1960 + "mov %0, %%"REG_S" \n\t"
1961 + ASMALIGN(4)
1962 + "1: \n\t"
1963 + "movlps (%%"REG_S"), %%xmm0\n\t" //buf[p]
1964 + "movlps 8(%%"REG_S"), %%xmm1\n\t" //buf[q]
1965 + "movhps (%%"REG_S"), %%xmm0\n\t" //buf[p]
1966 + "movhps 8(%%"REG_S"), %%xmm2\n\t" //buf[q]
1967 + "addps %%xmm1, %%xmm0 \n\t"
1968 + "subps %%xmm2, %%xmm0 \n\t"
1969 + "movaps %%xmm0, (%%"REG_S")\n\t"
1970 + "add $16, %%"REG_S" \n\t"
1971 + "cmp %1, %%"REG_S" \n\t"
1972 + " jb 1b \n\t"
1973 + :: "g" (buf), "r" (buf + 128)
1974 + : "%"REG_S
1975 + );
1976 +
1977 + /* 2. iteration */
1978 + // Note w[1]={{1,0}, {0,-1}}
1979 + __asm__ volatile(
1980 + "movaps "MANGLE(ps111_1)", %%xmm7\n\t" // 1,1,1,-1
1981 + "mov %0, %%"REG_S" \n\t"
1982 + ASMALIGN(4)
1983 + "1: \n\t"
1984 + "movaps 16(%%"REG_S"), %%xmm2 \n\t" //r2,i2,r3,i3
1985 + "shufps $0xB4, %%xmm2, %%xmm2 \n\t" //r2,i2,i3,r3
1986 + "mulps %%xmm7, %%xmm2 \n\t" //r2,i2,i3,-r3
1987 + "movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
1988 + "movaps (%%"REG_S"), %%xmm1 \n\t" //r0,i0,r1,i1
1989 + "addps %%xmm2, %%xmm0 \n\t"
1990 + "subps %%xmm2, %%xmm1 \n\t"
1991 + "movaps %%xmm0, (%%"REG_S") \n\t"
1992 + "movaps %%xmm1, 16(%%"REG_S") \n\t"
1993 + "add $32, %%"REG_S" \n\t"
1994 + "cmp %1, %%"REG_S" \n\t"
1995 + " jb 1b \n\t"
1996 + :: "g" (buf), "r" (buf + 128)
1997 + : "%"REG_S
1998 + );
1999 +
2000 + /* 3. iteration */
2001 +/*
2002 + Note sseW2+0={1,1,sqrt(2),sqrt(2))
2003 + Note sseW2+16={0,0,sqrt(2),-sqrt(2))
2004 + Note sseW2+32={0,0,-sqrt(2),-sqrt(2))
2005 + Note sseW2+48={1,-1,sqrt(2),-sqrt(2))
2006 +*/
2007 + __asm__ volatile(
2008 + "movaps 48+"MANGLE(sseW2)", %%xmm6\n\t"
2009 + "movaps 16+"MANGLE(sseW2)", %%xmm7\n\t"
2010 + "xorps %%xmm5, %%xmm5 \n\t"
2011 + "xorps %%xmm2, %%xmm2 \n\t"
2012 + "mov %0, %%"REG_S" \n\t"
2013 + ASMALIGN(4)
2014 + "1: \n\t"
2015 + "movaps 32(%%"REG_S"), %%xmm2 \n\t" //r4,i4,r5,i5
2016 + "movaps 48(%%"REG_S"), %%xmm3 \n\t" //r6,i6,r7,i7
2017 + "movaps "MANGLE(sseW2)", %%xmm4 \n\t" //r4,i4,r5,i5
2018 + "movaps 32+"MANGLE(sseW2)", %%xmm5\n\t" //r6,i6,r7,i7
2019 + "mulps %%xmm2, %%xmm4 \n\t"
2020 + "mulps %%xmm3, %%xmm5 \n\t"
2021 + "shufps $0xB1, %%xmm2, %%xmm2 \n\t" //i4,r4,i5,r5
2022 + "shufps $0xB1, %%xmm3, %%xmm3 \n\t" //i6,r6,i7,r7
2023 + "mulps %%xmm6, %%xmm3 \n\t"
2024 + "mulps %%xmm7, %%xmm2 \n\t"
2025 + "movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
2026 + "movaps 16(%%"REG_S"), %%xmm1 \n\t" //r2,i2,r3,i3
2027 + "addps %%xmm4, %%xmm2 \n\t"
2028 + "addps %%xmm5, %%xmm3 \n\t"
2029 + "movaps %%xmm2, %%xmm4 \n\t"
2030 + "movaps %%xmm3, %%xmm5 \n\t"
2031 + "addps %%xmm0, %%xmm2 \n\t"
2032 + "addps %%xmm1, %%xmm3 \n\t"
2033 + "subps %%xmm4, %%xmm0 \n\t"
2034 + "subps %%xmm5, %%xmm1 \n\t"
2035 + "movaps %%xmm2, (%%"REG_S") \n\t"
2036 + "movaps %%xmm3, 16(%%"REG_S") \n\t"
2037 + "movaps %%xmm0, 32(%%"REG_S") \n\t"
2038 + "movaps %%xmm1, 48(%%"REG_S") \n\t"
2039 + "add $64, %%"REG_S" \n\t"
2040 + "cmp %1, %%"REG_S" \n\t"
2041 + " jb 1b \n\t"
2042 + :: "g" (buf), "r" (buf + 128)
2043 + : "%"REG_S
2044 + );
2045 +
2046 + /* 4-7. iterations */
2047 + for (m=3; m < 7; m++) {
2048 + two_m = (1 << m);
2049 + two_m_plus_one = two_m<<1;
2050 + two_m_plus_one_shl3 = (two_m_plus_one<<3);
2051 + buf_offset = buf+128;
2052 + __asm__ volatile(
2053 + "mov %0, %%"REG_S" \n\t"
2054 + ASMALIGN(4)
2055 + "1: \n\t"
2056 + "xor %%"REG_D", %%"REG_D" \n\t" // k
2057 + "lea (%%"REG_S", %3), %%"REG_d" \n\t"
2058 + "2: \n\t"
2059 + "movaps (%%"REG_d", %%"REG_D"), %%xmm1 \n\t"
2060 + "movaps (%4, %%"REG_D", 2), %%xmm2 \n\t"
2061 + "mulps %%xmm1, %%xmm2 \n\t"
2062 + "shufps $0xB1, %%xmm1, %%xmm1 \n\t"
2063 + "mulps 16(%4, %%"REG_D", 2), %%xmm1 \n\t"
2064 + "movaps (%%"REG_S", %%"REG_D"), %%xmm0 \n\t"
2065 + "addps %%xmm2, %%xmm1 \n\t"
2066 + "movaps %%xmm1, %%xmm2 \n\t"
2067 + "addps %%xmm0, %%xmm1 \n\t"
2068 + "subps %%xmm2, %%xmm0 \n\t"
2069 + "movaps %%xmm1, (%%"REG_S", %%"REG_D") \n\t"
2070 + "movaps %%xmm0, (%%"REG_d", %%"REG_D") \n\t"
2071 + "add $16, %%"REG_D" \n\t"
2072 + "cmp %3, %%"REG_D" \n\t" //FIXME (opt) count against 0
2073 + "jb 2b \n\t"
2074 + "add %2, %%"REG_S" \n\t"
2075 + "cmp %1, %%"REG_S" \n\t"
2076 + " jb 1b \n\t"
2077 + :: "g" (buf), "m" (buf_offset), "m" (two_m_plus_one_shl3), "r" (two_m<<3),
2078 + "r" (sseW[m])
2079 + : "%"REG_S, "%"REG_D, "%"REG_d
2080 + );
2081 + }
2082 +
2083 + /* Post IFFT complex multiply plus IFFT complex conjugate*/
2084 + __asm__ volatile(
2085 + "mov $-1024, %%"REG_S" \n\t"
2086 + ASMALIGN(4)
2087 + "1: \n\t"
2088 + "movaps (%0, %%"REG_S"), %%xmm0 \n\t"
2089 + "movaps (%0, %%"REG_S"), %%xmm1 \n\t"
2090 + "shufps $0xB1, %%xmm0, %%xmm0 \n\t"
2091 + "mulps 1024+"MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm1\n\t"
2092 + "mulps 1024+"MANGLE(sseSinCos1d)"(%%"REG_S"), %%xmm0\n\t"
2093 + "addps %%xmm1, %%xmm0 \n\t"
2094 + "movaps %%xmm0, (%0, %%"REG_S") \n\t"
2095 + "add $16, %%"REG_S" \n\t"
2096 + " jnz 1b \n\t"
2097 + :: "r" (buf+128)
2098 + : "%"REG_S
2099 + );
2100 +
2101 +
2102 + data_ptr = data;
2103 + delay_ptr = delay;
2104 + window_ptr = a52_imdct_window;
2105 +
2106 + /* Window and convert to real valued signal */
2107 + __asm__ volatile(
2108 + "xor %%"REG_D", %%"REG_D" \n\t" // 0
2109 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2110 + "movss %3, %%xmm2 \n\t" // bias
2111 + "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
2112 + ASMALIGN(4)
2113 + "1: \n\t"
2114 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
2115 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
2116 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
2117 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // ? B A ?
2118 + "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
2119 + "mulps "MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2120 + "addps (%2, %%"REG_S"), %%xmm0 \n\t"
2121 + "addps %%xmm2, %%xmm0 \n\t"
2122 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2123 + "add $16, %%"REG_S" \n\t"
2124 + "sub $16, %%"REG_D" \n\t"
2125 + "cmp $512, %%"REG_S" \n\t"
2126 + " jb 1b \n\t"
2127 + :: "r" (buf+64), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
2128 + : "%"REG_S, "%"REG_D
2129 + );
2130 + data_ptr+=128;
2131 + delay_ptr+=128;
2132 +// window_ptr+=128;
2133 +
2134 + __asm__ volatile(
2135 + "mov $1024, %%"REG_D" \n\t" // 512
2136 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2137 + "movss %3, %%xmm2 \n\t" // bias
2138 + "shufps $0x00, %%xmm2, %%xmm2 \n\t" // bias, bias, ...
2139 + ASMALIGN(4)
2140 + "1: \n\t"
2141 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
2142 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
2143 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
2144 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // B ? ? A
2145 + "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
2146 + "mulps 512+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2147 + "addps (%2, %%"REG_S"), %%xmm0 \n\t"
2148 + "addps %%xmm2, %%xmm0 \n\t"
2149 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2150 + "add $16, %%"REG_S" \n\t"
2151 + "sub $16, %%"REG_D" \n\t"
2152 + "cmp $512, %%"REG_S" \n\t"
2153 + " jb 1b \n\t"
2154 + :: "r" (buf), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
2155 + : "%"REG_S, "%"REG_D
2156 + );
2157 + data_ptr+=128;
2158 +// window_ptr+=128;
2159 +
2160 + /* The trailing edge of the window goes into the delay line */
2161 + delay_ptr = delay;
2162 +
2163 + __asm__ volatile(
2164 + "xor %%"REG_D", %%"REG_D" \n\t" // 0
2165 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2166 + ASMALIGN(4)
2167 + "1: \n\t"
2168 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
2169 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
2170 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
2171 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // B ? ? A
2172 + "shufps $0xCC, %%xmm1, %%xmm0 \n\t" // D C B A
2173 + "mulps 1024+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2174 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2175 + "add $16, %%"REG_S" \n\t"
2176 + "sub $16, %%"REG_D" \n\t"
2177 + "cmp $512, %%"REG_S" \n\t"
2178 + " jb 1b \n\t"
2179 + :: "r" (buf+64), "r" (delay_ptr)
2180 + : "%"REG_S, "%"REG_D
2181 + );
2182 + delay_ptr+=128;
2183 +// window_ptr-=128;
2184 +
2185 + __asm__ volatile(
2186 + "mov $1024, %%"REG_D" \n\t" // 1024
2187 + "xor %%"REG_S", %%"REG_S" \n\t" // 0
2188 + ASMALIGN(4)
2189 + "1: \n\t"
2190 + "movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
2191 + "movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
2192 + "movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
2193 + "movhps -8(%0, %%"REG_D"), %%xmm0 \n\t" // ? B A ?
2194 + "shufps $0x99, %%xmm1, %%xmm0 \n\t" // D C B A
2195 + "mulps 1536+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
2196 + "movaps %%xmm0, (%1, %%"REG_S") \n\t"
2197 + "add $16, %%"REG_S" \n\t"
2198 + "sub $16, %%"REG_D" \n\t"
2199 + "cmp $512, %%"REG_S" \n\t"
2200 + " jb 1b \n\t"
2201 + :: "r" (buf), "r" (delay_ptr)
2202 + : "%"REG_S, "%"REG_D
2203 + );
2204 +}
2205 +#endif // ARCH_X86 || ARCH_X86_64
2206 +
2207 void a52_imdct_256(sample_t * data, sample_t * delay, sample_t bias)
2208 {
2209 int i, k;
2210 @@ -364,7 +1141,7 @@
2212 void a52_imdct_init (uint32_t mm_accel)
2213 {
2214 - int i, k;
2215 + int i, j, k;
2216 double sum;
2218 /* compute imdct window - kaiser-bessel derived window, alpha = 5.0 */
2219 @@ -416,6 +1193,99 @@
2220 post2[i].real = cos ((M_PI / 128) * (i + 0.5));
2221 post2[i].imag = sin ((M_PI / 128) * (i + 0.5));
2222 }
2223 + for (i = 0; i < 128; i++) {
2224 + xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));
2225 + xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
2226 + }
2227 + for (i = 0; i < 7; i++) {
2228 + j = 1 << i;
2229 + for (k = 0; k < j; k++) {
2230 + w[i][k].real = cos (-M_PI * k / j);
2231 + w[i][k].imag = sin (-M_PI * k / j);
2232 + }
2233 + }
2234 +#if ARCH_X86 || ARCH_X86_64
2235 + for (i = 0; i < 128; i++) {
2236 + sseSinCos1c[2*i+0]= xcos1[i];
2237 + sseSinCos1c[2*i+1]= -xcos1[i];
2238 + sseSinCos1d[2*i+0]= xsin1[i];
2239 + sseSinCos1d[2*i+1]= xsin1[i];
2240 + }
2241 + for (i = 1; i < 7; i++) {
2242 + j = 1 << i;
2243 + for (k = 0; k < j; k+=2) {
2244 +
2245 + sseW[i][4*k + 0] = w[i][k+0].real;
2246 + sseW[i][4*k + 1] = w[i][k+0].real;
2247 + sseW[i][4*k + 2] = w[i][k+1].real;
2248 + sseW[i][4*k + 3] = w[i][k+1].real;
2249 +
2250 + sseW[i][4*k + 4] = -w[i][k+0].imag;
2251 + sseW[i][4*k + 5] = w[i][k+0].imag;
2252 + sseW[i][4*k + 6] = -w[i][k+1].imag;
2253 + sseW[i][4*k + 7] = w[i][k+1].imag;
2254 +
2255 + //we multiply more or less uninitalized numbers so we need to use exactly 0.0
2256 + if(k==0)
2257 + {
2258 +// sseW[i][4*k + 0]= sseW[i][4*k + 1]= 1.0;
2259 + sseW[i][4*k + 4]= sseW[i][4*k + 5]= 0.0;
2260 + }
2261 +
2262 + if(2*k == j)
2263 + {
2264 + sseW[i][4*k + 0]= sseW[i][4*k + 1]= 0.0;
2265 +// sseW[i][4*k + 4]= -(sseW[i][4*k + 5]= -1.0);
2266 + }
2267 + }
2268 + }
2269 +
2270 + for(i=0; i<128; i++)
2271 + {
2272 + sseWindow[2*i+0]= -a52_imdct_window[2*i+0];
2273 + sseWindow[2*i+1]= a52_imdct_window[2*i+1];
2274 + }
2275 +
2276 + for(i=0; i<64; i++)
2277 + {
2278 + sseWindow[256 + 2*i+0]= -a52_imdct_window[254 - 2*i+1];
2279 + sseWindow[256 + 2*i+1]= a52_imdct_window[254 - 2*i+0];
2280 + sseWindow[384 + 2*i+0]= a52_imdct_window[126 - 2*i+1];
2281 + sseWindow[384 + 2*i+1]= -a52_imdct_window[126 - 2*i+0];
2282 + }
2283 +#endif
2284 + a52_imdct_512 = imdct_do_512;
2285 + ifft128 = ifft128_c;
2286 + ifft64 = ifft64_c;
2287 +
2288 +#if ARCH_X86 || ARCH_X86_64
2289 + if(mm_accel & MM_ACCEL_X86_SSE)
2290 + {
2291 + fprintf (stderr, "Using SSE optimized IMDCT transform\n");
2292 + a52_imdct_512 = imdct_do_512_sse;
2293 + }
2294 + else
2295 + if(mm_accel & MM_ACCEL_X86_3DNOWEXT)
2296 + {
2297 + fprintf (stderr, "Using 3DNowEx optimized IMDCT transform\n");
2298 + a52_imdct_512 = imdct_do_512_3dnowex;
2299 + }
2300 + else
2301 + if(mm_accel & MM_ACCEL_X86_3DNOW)
2302 + {
2303 + fprintf (stderr, "Using 3DNow optimized IMDCT transform\n");
2304 + a52_imdct_512 = imdct_do_512_3dnow;
2305 + }
2306 + else
2307 +#endif // ARCH_X86 || ARCH_X86_64
2308 +#if HAVE_ALTIVEC
2309 + if (mm_accel & MM_ACCEL_PPC_ALTIVEC)
2310 + {
2311 + fprintf(stderr, "Using AltiVec optimized IMDCT transform\n");
2312 + a52_imdct_512 = imdct_do_512_altivec;
2313 + }
2314 + else
2315 +#endif
2317 #ifdef LIBA52_DJBFFT
2318 if (mm_accel & MM_ACCEL_DJBFFT) {
2319 @@ -426,7 +1296,5 @@
2320 #endif
2321 {
2322 fprintf (stderr, "No accelerated IMDCT transform found\n");
2323 - ifft128 = ifft128_c;
2324 - ifft64 = ifft64_c;
2325 }
2326 }
2327 --- include/mm_accel.h 2006-06-12 15:05:00.000000000 +0200
2328 +++ liba52/mm_accel.h 2006-06-05 02:23:04.000000000 +0200
2329 @@ -30,7 +34,12 @@
2330 /* x86 accelerations */
2331 #define MM_ACCEL_X86_MMX 0x80000000
2332 #define MM_ACCEL_X86_3DNOW 0x40000000
2333 +#define MM_ACCEL_X86_3DNOWEXT 0x08000000
2334 #define MM_ACCEL_X86_MMXEXT 0x20000000
2335 +#define MM_ACCEL_X86_SSE 0x10000000
2336 +
2337 +/* PPC accelerations */
2338 +#define MM_ACCEL_PPC_ALTIVEC 0x00010000
2340 uint32_t mm_accel (void);
2342 --- liba52/parse.c 2006-12-05 08:08:01.000000000 +0100
2343 +++ liba52/parse.c 2006-12-05 08:08:44.000000000 +0100
2344 @@ -24,6 +28,7 @@
2345 #include "config.h"
2347 #include <stdlib.h>
2348 +#include <stdio.h>
2349 #include <string.h>
2350 #include <inttypes.h>
2352 @@ -31,13 +36,12 @@
2353 #include "a52_internal.h"
2354 #include "bitstream.h"
2355 #include "tables.h"
2356 +#include "mm_accel.h"
2357 +#include "libavutil/avutil.h"
2359 -#ifdef HAVE_MEMALIGN
2360 +#if HAVE_MEMALIGN
2361 /* some systems have memalign() but no declaration for it */
2362 void * memalign (size_t align, size_t size);
2363 -#else
2364 -/* assume malloc alignment is sufficient */
2365 -#define memalign(align,size) malloc (size)
2366 #endif
2368 typedef struct {
2369 @@ -60,7 +64,16 @@
2370 if (state == NULL)
2371 return NULL;
2373 +#if defined(__MINGW32__) && defined(HAVE_SSE)
2374 + state->samples = av_malloc(256 * 12 * sizeof (sample_t));
2375 +#else
2376 state->samples = memalign (16, 256 * 12 * sizeof (sample_t));
2377 +#endif
2378 + if(((int)state->samples%16) && (mm_accel&MM_ACCEL_X86_SSE)){
2379 + mm_accel &=~MM_ACCEL_X86_SSE;
2380 + fprintf(stderr, "liba52: unable to get 16 byte aligned memory disabling usage of SSE instructions\n");
2381 + }
2382 +
2383 if (state->samples == NULL) {
2384 free (state);
2385 return NULL;
2386 @@ -74,6 +87,7 @@
2387 state->lfsr_state = 1;
2389 a52_imdct_init (mm_accel);
2390 + downmix_accel_init(mm_accel);
2392 return state;
2393 }
2394 @@ -141,7 +155,7 @@
2395 state->acmod = acmod = buf[6] >> 5;
2397 a52_bitstream_set_ptr (state, buf + 6);
2398 - bitstream_get (state, 3); /* skip acmod we already parsed */
2399 + bitstream_skip (state, 3); /* skip acmod we already parsed */
2401 if ((acmod == 2) && (bitstream_get (state, 2) == 2)) /* dsurmod */
2402 acmod = A52_DOLBY;
2403 @@ -172,28 +186,28 @@
2405 chaninfo = !acmod;
2406 do {
2407 - bitstream_get (state, 5); /* dialnorm */
2408 + bitstream_skip (state, 5); /* dialnorm */
2409 if (bitstream_get (state, 1)) /* compre */
2410 - bitstream_get (state, 8); /* compr */
2411 + bitstream_skip (state, 8); /* compr */
2412 if (bitstream_get (state, 1)) /* langcode */
2413 - bitstream_get (state, 8); /* langcod */
2414 + bitstream_skip (state, 8); /* langcod */
2415 if (bitstream_get (state, 1)) /* audprodie */
2416 - bitstream_get (state, 7); /* mixlevel + roomtyp */
2417 + bitstream_skip (state, 7); /* mixlevel + roomtyp */
2418 } while (chaninfo--);
2420 - bitstream_get (state, 2); /* copyrightb + origbs */
2421 + bitstream_skip (state, 2); /* copyrightb + origbs */
2423 if (bitstream_get (state, 1)) /* timecod1e */
2424 - bitstream_get (state, 14); /* timecod1 */
2425 + bitstream_skip (state, 14); /* timecod1 */
2426 if (bitstream_get (state, 1)) /* timecod2e */
2427 - bitstream_get (state, 14); /* timecod2 */
2428 + bitstream_skip (state, 14); /* timecod2 */
2430 if (bitstream_get (state, 1)) { /* addbsie */
2431 int addbsil;
2433 addbsil = bitstream_get (state, 6);
2434 do {
2435 - bitstream_get (state, 8); /* addbsi */
2436 + bitstream_skip (state, 8); /* addbsi */
2437 } while (addbsil--);
2438 }
2440 @@ -680,7 +694,7 @@
2441 state->fbw_expbap[i].exp[0],
2442 state->fbw_expbap[i].exp + 1))
2443 return 1;
2444 - bitstream_get (state, 2); /* gainrng */
2445 + bitstream_skip (state, 2); /* gainrng */
2446 }
2447 if (lfeexpstr != EXP_REUSE) {
2448 do_bit_alloc |= 32;
2449 @@ -755,7 +769,7 @@
2450 if (bitstream_get (state, 1)) { /* skiple */
2451 i = bitstream_get (state, 9); /* skipl */
2452 while (i--)
2453 - bitstream_get (state, 8);
2454 + bitstream_skip (state, 8);
2455 }
2457 samples = state->samples;
2458 @@ -896,6 +910,10 @@
2460 void a52_free (a52_state_t * state)
2461 {
2462 - free (state->samples);
2463 +#if defined(__MINGW32__) && defined(HAVE_SSE)
2464 + av_free (state->samples);
2465 +#else
2466 + free (state->samples);
2467 +#endif
2468 free (state);
2469 }