Mercurial > mplayer.hg

comparison liba52/resample_mmx.c @ 3909:ef32c8bdee81

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
c, mmx versions separated. a52 style runtime stuff
author arpi
date Sun, 30 Dec 2001 21:44:10 +0000
parents 0cc94b1eec0f
children 2dbd637ffe05
comparison
equal deleted inserted replaced
3908:0cc94b1eec0f 3909:ef32c8bdee81
1 // this code come from a52dec/libao/audio_out_oss.c
2
3 // FIXME FIXME FIXME
4
5 // a52_resample_init should find the requested converter (from type flags ->
6 // given number of channels) and set up some function pointers...
7
8 // a52_resample() should do the conversion.
9 1
10 // MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL) 2 // MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
11 3
12 /* optimization TODO / NOTES 4 /* optimization TODO / NOTES
13 movntq is slightly faster (0.5% with the current test.c benchmark) 5 movntq is slightly faster (0.5% with the current test.c benchmark)
14 (but thats just test.c so that needs to be testd in reallity) 6 (but thats just test.c so that needs to be testd in reallity)
15 and it would mean (C / MMX2 / MMX / 3DNOW) versions 7 and it would mean (C / MMX2 / MMX / 3DNOW) versions
16 */ 8 */
17 9
18 #include <inttypes.h>
19 #include <stdio.h>
20 #include "a52.h"
21 #include "mm_accel.h"
22 #include "../config.h"
23
24 int (* a52_resample) (float * _f, int16_t * s16)=NULL;
25
26 #ifdef ARCH_X86
27 static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL; 10 static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL;
28 static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL; 11 static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL;
29 static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL; 12 static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL;
30 static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL; 13 static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL;
31 #endif 14
32 15 static int a52_resample_MONO_to_5_MMX(float * _f, int16_t * s16){
33 static inline int16_t convert (int32_t i) 16 int i;
34 { 17 int32_t * f = (int32_t *) _f;
35 if (i > 0x43c07fff)
36 return 32767;
37 else if (i < 0x43bf8000)
38 return -32768;
39 else
40 return i - 0x43c00000;
41 }
42
43 static int chans=2;
44 static int flags=0;
45
46 int a52_resample_C(float * _f, int16_t * s16)
47 {
48 int i;
49 int32_t * f = (int32_t *) _f;
50
51 switch (flags) {
52 case A52_MONO:
53 for (i = 0; i < 256; i++) {
54 s16[5*i] = s16[5*i+1] = s16[5*i+2] = s16[5*i+3] = 0;
55 s16[5*i+4] = convert (f[i]);
56 }
57 break;
58 case A52_CHANNEL:
59 case A52_STEREO:
60 case A52_DOLBY:
61 for (i = 0; i < 256; i++) {
62 s16[2*i] = convert (f[i]);
63 s16[2*i+1] = convert (f[i+256]);
64 }
65 break;
66 case A52_3F:
67 for (i = 0; i < 256; i++) {
68 s16[5*i] = convert (f[i]);
69 s16[5*i+1] = convert (f[i+512]);
70 s16[5*i+2] = s16[5*i+3] = 0;
71 s16[5*i+4] = convert (f[i+256]);
72 }
73 break;
74 case A52_2F2R:
75 for (i = 0; i < 256; i++) {
76 s16[4*i] = convert (f[i]);
77 s16[4*i+1] = convert (f[i+256]);
78 s16[4*i+2] = convert (f[i+512]);
79 s16[4*i+3] = convert (f[i+768]);
80 }
81 break;
82 case A52_3F2R:
83 for (i = 0; i < 256; i++) {
84 s16[5*i] = convert (f[i]);
85 s16[5*i+1] = convert (f[i+512]);
86 s16[5*i+2] = convert (f[i+768]);
87 s16[5*i+3] = convert (f[i+1024]);
88 s16[5*i+4] = convert (f[i+256]);
89 }
90 break;
91 case A52_MONO | A52_LFE:
92 for (i = 0; i < 256; i++) {
93 s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
94 s16[6*i+4] = convert (f[i+256]);
95 s16[6*i+5] = convert (f[i]);
96 }
97 break;
98 case A52_CHANNEL | A52_LFE:
99 case A52_STEREO | A52_LFE:
100 case A52_DOLBY | A52_LFE:
101 for (i = 0; i < 256; i++) {
102 s16[6*i] = convert (f[i+256]);
103 s16[6*i+1] = convert (f[i+512]);
104 s16[6*i+2] = s16[6*i+3] = s16[6*i+4] = 0;
105 s16[6*i+5] = convert (f[i]);
106 }
107 break;
108 case A52_3F | A52_LFE:
109 for (i = 0; i < 256; i++) {
110 s16[6*i] = convert (f[i+256]);
111 s16[6*i+1] = convert (f[i+768]);
112 s16[6*i+2] = s16[6*i+3] = 0;
113 s16[6*i+4] = convert (f[i+512]);
114 s16[6*i+5] = convert (f[i]);
115 }
116 break;
117 case A52_2F2R | A52_LFE:
118 for (i = 0; i < 256; i++) {
119 s16[6*i] = convert (f[i+256]);
120 s16[6*i+1] = convert (f[i+512]);
121 s16[6*i+2] = convert (f[i+768]);
122 s16[6*i+3] = convert (f[i+1024]);
123 s16[6*i+4] = 0;
124 s16[6*i+5] = convert (f[i]);
125 }
126 break;
127 case A52_3F2R | A52_LFE:
128 for (i = 0; i < 256; i++) {
129 s16[6*i] = convert (f[i+256]);
130 s16[6*i+1] = convert (f[i+768]);
131 s16[6*i+2] = convert (f[i+1024]);
132 s16[6*i+3] = convert (f[i+1280]);
133 s16[6*i+4] = convert (f[i+512]);
134 s16[6*i+5] = convert (f[i]);
135 }
136 break;
137 }
138 return chans*256;
139 }
140
141 #ifdef ARCH_X86
142 int a52_resample_MMX(float * _f, int16_t * s16)
143 {
144 int i;
145 int32_t * f = (int32_t *) _f;
146
147 switch (flags) {
148 case A52_MONO:
149 asm volatile( 18 asm volatile(
150 "movl $-512, %%esi \n\t" 19 "movl $-512, %%esi \n\t"
151 "movq magicF2W, %%mm7 \n\t" 20 "movq magicF2W, %%mm7 \n\t"
152 "movq wm1100, %%mm3 \n\t" 21 "movq wm1100, %%mm3 \n\t"
153 "movq wm0101, %%mm4 \n\t" 22 "movq wm0101, %%mm4 \n\t"
176 " jnz 1b \n\t" 45 " jnz 1b \n\t"
177 "emms \n\t" 46 "emms \n\t"
178 :: "r" (s16+1280), "r" (f+256) 47 :: "r" (s16+1280), "r" (f+256)
179 :"%esi", "%edi", "memory" 48 :"%esi", "%edi", "memory"
180 ); 49 );
181 break; 50 return 5*256;
182 case A52_CHANNEL: 51 }
183 case A52_STEREO: 52
184 case A52_DOLBY: 53 static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){
54 int i;
55 int32_t * f = (int32_t *) _f;
185 /* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it 56 /* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it
186 #ifdef HAVE_SSE 57 #ifdef HAVE_SSE
187 asm volatile( 58 asm volatile(
188 "movl $-1024, %%esi \n\t" 59 "movl $-1024, %%esi \n\t"
189 "1: \n\t" 60 "1: \n\t"
223 " jnz 1b \n\t" 94 " jnz 1b \n\t"
224 "emms \n\t" 95 "emms \n\t"
225 :: "r" (s16+512), "r" (f+256) 96 :: "r" (s16+512), "r" (f+256)
226 :"%esi", "memory" 97 :"%esi", "memory"
227 ); 98 );
228 break; 99 return 2*256;
229 case A52_3F: 100 }
101
102 static int a52_resample_3F_to_5_MMX(float * _f, int16_t * s16){
103 int i;
104 int32_t * f = (int32_t *) _f;
230 asm volatile( 105 asm volatile(
231 "movl $-1024, %%esi \n\t" 106 "movl $-1024, %%esi \n\t"
232 "movq magicF2W, %%mm7 \n\t" 107 "movq magicF2W, %%mm7 \n\t"
233 "pxor %%mm6, %%mm6 \n\t" 108 "pxor %%mm6, %%mm6 \n\t"
234 "movq %%mm7, %%mm5 \n\t" 109 "movq %%mm7, %%mm5 \n\t"
275 " jnz 1b \n\t" 150 " jnz 1b \n\t"
276 "emms \n\t" 151 "emms \n\t"
277 :: "r" (s16+1280), "r" (f+256) 152 :: "r" (s16+1280), "r" (f+256)
278 :"%esi", "%edi", "memory" 153 :"%esi", "%edi", "memory"
279 ); 154 );
280 break; 155 return 5*256;
281 case A52_2F2R: 156 }
157
158 static int a52_resample_2F_2R_to_4_MMX(float * _f, int16_t * s16){
159 int i;
160 int32_t * f = (int32_t *) _f;
282 asm volatile( 161 asm volatile(
283 "movl $-1024, %%esi \n\t" 162 "movl $-1024, %%esi \n\t"
284 "movq magicF2W, %%mm7 \n\t" 163 "movq magicF2W, %%mm7 \n\t"
285 "1: \n\t" 164 "1: \n\t"
286 "movq (%1, %%esi), %%mm0 \n\t" 165 "movq (%1, %%esi), %%mm0 \n\t"
323 " jnz 1b \n\t" 202 " jnz 1b \n\t"
324 "emms \n\t" 203 "emms \n\t"
325 :: "r" (s16+1024), "r" (f+256) 204 :: "r" (s16+1024), "r" (f+256)
326 :"%esi", "memory" 205 :"%esi", "memory"
327 ); 206 );
328 break; 207 return 4*256;
329 case A52_3F2R: 208 }
209
210 static int a52_resample_3F_2R_to_5_MMX(float * _f, int16_t * s16){
211 int i;
212 int32_t * f = (int32_t *) _f;
330 asm volatile( 213 asm volatile(
331 "movl $-1024, %%esi \n\t" 214 "movl $-1024, %%esi \n\t"
332 "movq magicF2W, %%mm7 \n\t" 215 "movq magicF2W, %%mm7 \n\t"
333 "1: \n\t" 216 "1: \n\t"
334 "movd (%1, %%esi), %%mm0 \n\t" 217 "movd (%1, %%esi), %%mm0 \n\t"
379 " jnz 1b \n\t" 262 " jnz 1b \n\t"
380 "emms \n\t" 263 "emms \n\t"
381 :: "r" (s16+1280), "r" (f+256) 264 :: "r" (s16+1280), "r" (f+256)
382 :"%esi", "%edi", "memory" 265 :"%esi", "%edi", "memory"
383 ); 266 );
384 break; 267 return 5*256;
385 case A52_MONO | A52_LFE: 268 }
269
270 static int a52_resample_MONO_LFE_to_6_MMX(float * _f, int16_t * s16){
271 int i;
272 int32_t * f = (int32_t *) _f;
386 asm volatile( 273 asm volatile(
387 "movl $-1024, %%esi \n\t" 274 "movl $-1024, %%esi \n\t"
388 "movq magicF2W, %%mm7 \n\t" 275 "movq magicF2W, %%mm7 \n\t"
389 "pxor %%mm6, %%mm6 \n\t" 276 "pxor %%mm6, %%mm6 \n\t"
390 "1: \n\t" 277 "1: \n\t"
416 " jnz 1b \n\t" 303 " jnz 1b \n\t"
417 "emms \n\t" 304 "emms \n\t"
418 :: "r" (s16+1536), "r" (f+256) 305 :: "r" (s16+1536), "r" (f+256)
419 :"%esi", "%edi", "memory" 306 :"%esi", "%edi", "memory"
420 ); 307 );
421 break; 308 return 6*256;
422 case A52_CHANNEL | A52_LFE: 309 }
423 case A52_STEREO | A52_LFE: 310
424 case A52_DOLBY | A52_LFE: 311 static int a52_resample_STEREO_LFE_to_6_MMX(float * _f, int16_t * s16){
312 int i;
313 int32_t * f = (int32_t *) _f;
425 asm volatile( 314 asm volatile(
426 "movl $-1024, %%esi \n\t" 315 "movl $-1024, %%esi \n\t"
427 "movq magicF2W, %%mm7 \n\t" 316 "movq magicF2W, %%mm7 \n\t"
428 "pxor %%mm6, %%mm6 \n\t" 317 "pxor %%mm6, %%mm6 \n\t"
429 "1: \n\t" 318 "1: \n\t"
453 " jnz 1b \n\t" 342 " jnz 1b \n\t"
454 "emms \n\t" 343 "emms \n\t"
455 :: "r" (s16+1536), "r" (f+256) 344 :: "r" (s16+1536), "r" (f+256)
456 :"%esi", "%edi", "memory" 345 :"%esi", "%edi", "memory"
457 ); 346 );
458 break; 347 return 6*256;
459 case A52_3F | A52_LFE: 348 }
349
350 static int a52_resample_3F_LFE_to_6_MMX(float * _f, int16_t * s16){
351 int i;
352 int32_t * f = (int32_t *) _f;
460 asm volatile( 353 asm volatile(
461 "movl $-1024, %%esi \n\t" 354 "movl $-1024, %%esi \n\t"
462 "movq magicF2W, %%mm7 \n\t" 355 "movq magicF2W, %%mm7 \n\t"
463 "pxor %%mm6, %%mm6 \n\t" 356 "pxor %%mm6, %%mm6 \n\t"
464 "1: \n\t" 357 "1: \n\t"
490 " jnz 1b \n\t" 383 " jnz 1b \n\t"
491 "emms \n\t" 384 "emms \n\t"
492 :: "r" (s16+1536), "r" (f+256) 385 :: "r" (s16+1536), "r" (f+256)
493 :"%esi", "%edi", "memory" 386 :"%esi", "%edi", "memory"
494 ); 387 );
495 break; 388 return 6*256;
496 case A52_2F2R | A52_LFE: 389 }
390
391 static int a52_resample_2F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
392 int i;
393 int32_t * f = (int32_t *) _f;
497 asm volatile( 394 asm volatile(
498 "movl $-1024, %%esi \n\t" 395 "movl $-1024, %%esi \n\t"
499 "movq magicF2W, %%mm7 \n\t" 396 "movq magicF2W, %%mm7 \n\t"
500 // "pxor %%mm6, %%mm6 \n\t" 397 // "pxor %%mm6, %%mm6 \n\t"
501 "1: \n\t" 398 "1: \n\t"
533 " jnz 1b \n\t" 430 " jnz 1b \n\t"
534 "emms \n\t" 431 "emms \n\t"
535 :: "r" (s16+1536), "r" (f+256) 432 :: "r" (s16+1536), "r" (f+256)
536 :"%esi", "%edi", "memory" 433 :"%esi", "%edi", "memory"
537 ); 434 );
538 break; 435 return 6*256;
539 case A52_3F2R | A52_LFE: 436 }
437
438 static int a52_resample_3F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
439 int i;
440 int32_t * f = (int32_t *) _f;
540 asm volatile( 441 asm volatile(
541 "movl $-1024, %%esi \n\t" 442 "movl $-1024, %%esi \n\t"
542 "movq magicF2W, %%mm7 \n\t" 443 "movq magicF2W, %%mm7 \n\t"
543 // "pxor %%mm6, %%mm6 \n\t" 444 // "pxor %%mm6, %%mm6 \n\t"
544 "1: \n\t" 445 "1: \n\t"
578 " jnz 1b \n\t" 479 " jnz 1b \n\t"
579 "emms \n\t" 480 "emms \n\t"
580 :: "r" (s16+1536), "r" (f+256) 481 :: "r" (s16+1536), "r" (f+256)
581 :"%esi", "%edi", "memory" 482 :"%esi", "%edi", "memory"
582 ); 483 );
484 return 6*256;
485 }
486
487
488 static void* a52_resample_MMX(int flags, int ch){
489 switch (flags) {
490 case A52_MONO:
491 if(ch==5) return a52_resample_MONO_to_5_MMX;
492 break;
493 case A52_CHANNEL:
494 case A52_STEREO:
495 case A52_DOLBY:
496 if(ch==2) return a52_resample_STEREO_to_2_MMX;
497 break;
498 case A52_3F:
499 if(ch==5) return a52_resample_3F_to_5_MMX;
500 break;
501 case A52_2F2R:
502 if(ch==4) return a52_resample_2F_2R_to_4_MMX;
503 break;
504 case A52_3F2R:
505 if(ch==5) return a52_resample_3F_2R_to_5_MMX;
506 break;
507 case A52_MONO | A52_LFE:
508 if(ch==6) return a52_resample_MONO_LFE_to_6_MMX;
509 break;
510 case A52_CHANNEL | A52_LFE:
511 case A52_STEREO | A52_LFE:
512 case A52_DOLBY | A52_LFE:
513 if(ch==6) return a52_resample_STEREO_LFE_to_6_MMX;
514 break;
515 case A52_3F | A52_LFE:
516 if(ch==6) return a52_resample_3F_LFE_to_6_MMX;
517 break;
518 case A52_2F2R | A52_LFE:
519 if(ch==6) return a52_resample_2F_2R_LFE_to_6_MMX;
520 break;
521 case A52_3F2R | A52_LFE:
522 if(ch==6) return a52_resample_3F_2R_LFE_to_6_MMX;
583 break; 523 break;
584 } 524 }
585 return chans*256; 525 return NULL;
586 } 526 }
587 #endif //arch_x86 527
588 528
589 void a52_resample_init(uint32_t mm_accel,int _flags,int _chans){
590 chans=_chans;
591 flags=_flags;
592
593 if(a52_resample==NULL) // only once please ;)
594 {
595 if(mm_accel & MM_ACCEL_X86_MMX) fprintf(stderr, "Using MMX optimized resampler\n");
596 else fprintf(stderr, "No accelerated resampler found\n");
597 }
598
599 #ifdef ARCH_X86
600 if(mm_accel & MM_ACCEL_X86_MMX) a52_resample= a52_resample_MMX;
601 #else
602 if(0);
603 #endif
604 else a52_resample= a52_resample_C;
605 }
606