Mercurial > mplayer.hg
comparison liba52/resample_mmx.c @ 3567:9e1e88b3ca18
mmx opt
| author | michael |
|---|---|
| date | Mon, 17 Dec 2001 19:33:19 +0000 |
| parents | 21d65a4ae3c9 |
| children | d14494d84c29 |
comparison
equal
deleted
inserted
replaced
| 3566:92b3028b10f3 | 3567:9e1e88b3ca18 |
|---|---|
| 7 | 7 |
| 8 // a52_resample() should do the conversion. | 8 // a52_resample() should do the conversion. |
| 9 | 9 |
| 10 #include <inttypes.h> | 10 #include <inttypes.h> |
| 11 #include "a52.h" | 11 #include "a52.h" |
| 12 #include "../config.h" | |
| 13 | |
| 14 #ifdef HAVE_MMX | |
| 15 static uint64_t __attribute__((aligned(16))) magicF2W= 0x43c0000043c00000LL; | |
| 16 #endif | |
| 12 | 17 |
| 13 static inline int16_t convert (int32_t i) | 18 static inline int16_t convert (int32_t i) |
| 14 { | 19 { |
| 15 if (i > 0x43c07fff) | 20 if (i > 0x43c07fff) |
| 16 return 32767; | 21 return 32767; |
| 41 } | 46 } |
| 42 break; | 47 break; |
| 43 case A52_CHANNEL: | 48 case A52_CHANNEL: |
| 44 case A52_STEREO: | 49 case A52_STEREO: |
| 45 case A52_DOLBY: | 50 case A52_DOLBY: |
| 51 /* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it | |
| 52 #ifdef HAVE_SSE | |
| 53 asm volatile( | |
| 54 "movl $-1024, %%esi \n\t" | |
| 55 "1: \n\t" | |
| 56 "cvtps2pi (%1, %%esi), %%mm0 \n\t" | |
| 57 "cvtps2pi 1024(%1, %%esi), %%mm2\n\t" | |
| 58 "movq %%mm0, %%mm1 \n\t" | |
| 59 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 60 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 61 "movq %%mm0, (%0, %%esi) \n\t" | |
| 62 "movq %%mm1, 8(%0, %%esi) \n\t" | |
| 63 "addl $16, %%esi \n\t" | |
| 64 " jnz 1b \n\t" | |
| 65 "emms \n\t" | |
| 66 :: "r" (s16+512), "r" (f+256) | |
| 67 :"%esi", "memory" | |
| 68 );*/ | |
| 69 #ifdef HAVE_MMX | |
| 70 asm volatile( | |
| 71 "movl $-1024, %%esi \n\t" | |
| 72 "movq magicF2W, %%mm7 \n\t" | |
| 73 "1: \n\t" | |
| 74 "movq (%1, %%esi), %%mm0 \n\t" | |
| 75 "movq 8(%1, %%esi), %%mm1 \n\t" | |
| 76 "movq 1024(%1, %%esi), %%mm2 \n\t" | |
| 77 "movq 1032(%1, %%esi), %%mm3 \n\t" | |
| 78 "psubd %%mm7, %%mm0 \n\t" | |
| 79 "psubd %%mm7, %%mm1 \n\t" | |
| 80 "psubd %%mm7, %%mm2 \n\t" | |
| 81 "psubd %%mm7, %%mm3 \n\t" | |
| 82 "packssdw %%mm1, %%mm0 \n\t" | |
| 83 "packssdw %%mm3, %%mm2 \n\t" | |
| 84 "movq %%mm0, %%mm1 \n\t" | |
| 85 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 86 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 87 "movq %%mm0, (%0, %%esi) \n\t" | |
| 88 "movq %%mm1, 8(%0, %%esi) \n\t" | |
| 89 "addl $16, %%esi \n\t" | |
| 90 " jnz 1b \n\t" | |
| 91 "emms \n\t" | |
| 92 :: "r" (s16+512), "r" (f+256) | |
| 93 :"%esi", "memory" | |
| 94 ); | |
| 95 #else | |
| 46 for (i = 0; i < 256; i++) { | 96 for (i = 0; i < 256; i++) { |
| 47 s16[2*i] = convert (f[i]); | 97 s16[2*i] = convert (f[i]); |
| 48 s16[2*i+1] = convert (f[i+256]); | 98 s16[2*i+1] = convert (f[i+256]); |
| 49 } | 99 } |
| 100 #endif | |
| 50 break; | 101 break; |
| 51 case A52_3F: | 102 case A52_3F: |
| 52 for (i = 0; i < 256; i++) { | 103 for (i = 0; i < 256; i++) { |
| 53 s16[5*i] = convert (f[i]); | 104 s16[5*i] = convert (f[i]); |
| 54 s16[5*i+1] = convert (f[i+512]); | 105 s16[5*i+1] = convert (f[i+512]); |