Mercurial > mplayer.hg
annotate liba52/resample_mmx.c @ 4112:3822582553d6
Alpha patch by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
| author | nick |
|---|---|
| date | Sat, 12 Jan 2002 18:00:25 +0000 |
| parents | ef32c8bdee81 |
| children | 2dbd637ffe05 |
| rev | line source |
|---|---|
|
3412
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
1 |
| 3569 | 2 // MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL) |
| 3 | |
| 4 /* optimization TODO / NOTES | |
| 5 movntq is slightly faster (0.5% with the current test.c benchmark) | |
| 6 (but thats just test.c so that needs to be testd in reallity) | |
| 7 and it would mean (C / MMX2 / MMX / 3DNOW) versions | |
| 8 */ | |
| 9 | |
| 3574 | 10 static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL; |
| 11 static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL; | |
| 12 static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL; | |
| 13 static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL; | |
|
3412
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
14 |
| 3909 | 15 static int a52_resample_MONO_to_5_MMX(float * _f, int16_t * s16){ |
|
3412
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
16 int i; |
|
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
17 int32_t * f = (int32_t *) _f; |
| 3574 | 18 asm volatile( |
| 19 "movl $-512, %%esi \n\t" | |
| 20 "movq magicF2W, %%mm7 \n\t" | |
| 21 "movq wm1100, %%mm3 \n\t" | |
| 22 "movq wm0101, %%mm4 \n\t" | |
| 23 "movq wm1010, %%mm5 \n\t" | |
| 24 "pxor %%mm6, %%mm6 \n\t" | |
| 25 "1: \n\t" | |
| 26 "movq (%1, %%esi, 2), %%mm0 \n\t" | |
| 27 "movq 8(%1, %%esi, 2), %%mm1 \n\t" | |
| 28 "leal (%%esi, %%esi, 4), %%edi \n\t" | |
| 29 "psubd %%mm7, %%mm0 \n\t" | |
| 30 "psubd %%mm7, %%mm1 \n\t" | |
| 31 "packssdw %%mm1, %%mm0 \n\t" | |
| 32 "movq %%mm0, %%mm1 \n\t" | |
| 33 "pand %%mm4, %%mm0 \n\t" | |
| 34 "pand %%mm5, %%mm1 \n\t" | |
| 35 "movq %%mm6, (%0, %%edi) \n\t" // 0 0 0 0 | |
| 36 "movd %%mm0, 8(%0, %%edi) \n\t" // A 0 | |
| 37 "pand %%mm3, %%mm0 \n\t" | |
| 38 "movd %%mm6, 12(%0, %%edi) \n\t" // 0 0 | |
| 39 "movd %%mm1, 16(%0, %%edi) \n\t" // 0 B | |
| 40 "pand %%mm3, %%mm1 \n\t" | |
| 41 "movd %%mm6, 20(%0, %%edi) \n\t" // 0 0 | |
| 42 "movq %%mm0, 24(%0, %%edi) \n\t" // 0 0 C 0 | |
| 43 "movq %%mm1, 32(%0, %%edi) \n\t" // 0 0 0 B | |
| 44 "addl $8, %%esi \n\t" | |
| 45 " jnz 1b \n\t" | |
| 46 "emms \n\t" | |
| 47 :: "r" (s16+1280), "r" (f+256) | |
| 48 :"%esi", "%edi", "memory" | |
| 49 ); | |
| 3909 | 50 return 5*256; |
| 51 } | |
| 52 | |
| 53 static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){ | |
| 54 int i; | |
| 55 int32_t * f = (int32_t *) _f; | |
| 3567 | 56 /* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it |
| 57 #ifdef HAVE_SSE | |
| 58 asm volatile( | |
| 59 "movl $-1024, %%esi \n\t" | |
| 60 "1: \n\t" | |
| 61 "cvtps2pi (%1, %%esi), %%mm0 \n\t" | |
| 62 "cvtps2pi 1024(%1, %%esi), %%mm2\n\t" | |
| 63 "movq %%mm0, %%mm1 \n\t" | |
| 64 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 65 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 66 "movq %%mm0, (%0, %%esi) \n\t" | |
| 67 "movq %%mm1, 8(%0, %%esi) \n\t" | |
| 68 "addl $16, %%esi \n\t" | |
| 69 " jnz 1b \n\t" | |
| 70 "emms \n\t" | |
| 71 :: "r" (s16+512), "r" (f+256) | |
| 72 :"%esi", "memory" | |
| 73 );*/ | |
| 74 asm volatile( | |
| 75 "movl $-1024, %%esi \n\t" | |
| 76 "movq magicF2W, %%mm7 \n\t" | |
| 77 "1: \n\t" | |
| 78 "movq (%1, %%esi), %%mm0 \n\t" | |
| 79 "movq 8(%1, %%esi), %%mm1 \n\t" | |
| 80 "movq 1024(%1, %%esi), %%mm2 \n\t" | |
| 81 "movq 1032(%1, %%esi), %%mm3 \n\t" | |
| 82 "psubd %%mm7, %%mm0 \n\t" | |
| 83 "psubd %%mm7, %%mm1 \n\t" | |
| 84 "psubd %%mm7, %%mm2 \n\t" | |
| 85 "psubd %%mm7, %%mm3 \n\t" | |
| 86 "packssdw %%mm1, %%mm0 \n\t" | |
| 87 "packssdw %%mm3, %%mm2 \n\t" | |
| 88 "movq %%mm0, %%mm1 \n\t" | |
| 89 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 90 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 91 "movq %%mm0, (%0, %%esi) \n\t" | |
| 92 "movq %%mm1, 8(%0, %%esi) \n\t" | |
| 93 "addl $16, %%esi \n\t" | |
| 94 " jnz 1b \n\t" | |
| 95 "emms \n\t" | |
| 96 :: "r" (s16+512), "r" (f+256) | |
| 97 :"%esi", "memory" | |
| 98 ); | |
| 3909 | 99 return 2*256; |
| 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; | |
| 3654 | 105 asm volatile( |
| 106 "movl $-1024, %%esi \n\t" | |
| 107 "movq magicF2W, %%mm7 \n\t" | |
| 108 "pxor %%mm6, %%mm6 \n\t" | |
| 109 "movq %%mm7, %%mm5 \n\t" | |
| 110 "punpckldq %%mm6, %%mm5 \n\t" | |
| 111 "1: \n\t" | |
| 112 "movd (%1, %%esi), %%mm0 \n\t" | |
| 113 "punpckldq 2048(%1, %%esi), %%mm0\n\t" | |
| 114 "movd 1024(%1, %%esi), %%mm1 \n\t" | |
| 115 "punpckldq 4(%1, %%esi), %%mm1 \n\t" | |
| 116 "movd 2052(%1, %%esi), %%mm2 \n\t" | |
| 117 "movq %%mm7, %%mm3 \n\t" | |
| 118 "punpckldq 1028(%1, %%esi), %%mm3\n\t" | |
| 119 "movd 8(%1, %%esi), %%mm4 \n\t" | |
| 120 "punpckldq 2056(%1, %%esi), %%mm4\n\t" | |
| 121 "leal (%%esi, %%esi, 4), %%edi \n\t" | |
| 122 "sarl $1, %%edi \n\t" | |
| 123 "psubd %%mm7, %%mm0 \n\t" | |
| 124 "psubd %%mm7, %%mm1 \n\t" | |
| 125 "psubd %%mm5, %%mm2 \n\t" | |
| 126 "psubd %%mm7, %%mm3 \n\t" | |
| 127 "psubd %%mm7, %%mm4 \n\t" | |
| 128 "packssdw %%mm6, %%mm0 \n\t" | |
| 129 "packssdw %%mm2, %%mm1 \n\t" | |
| 130 "packssdw %%mm4, %%mm3 \n\t" | |
| 131 "movq %%mm0, (%0, %%edi) \n\t" | |
| 132 "movq %%mm1, 8(%0, %%edi) \n\t" | |
| 133 "movq %%mm3, 16(%0, %%edi) \n\t" | |
| 134 | |
| 135 "movd 1032(%1, %%esi), %%mm1 \n\t" | |
| 136 "punpckldq 12(%1, %%esi), %%mm1\n\t" | |
| 137 "movd 2060(%1, %%esi), %%mm2 \n\t" | |
| 138 "movq %%mm7, %%mm3 \n\t" | |
| 139 "punpckldq 1036(%1, %%esi), %%mm3\n\t" | |
| 140 "pxor %%mm0, %%mm0 \n\t" | |
| 141 "psubd %%mm7, %%mm1 \n\t" | |
| 142 "psubd %%mm5, %%mm2 \n\t" | |
| 143 "psubd %%mm7, %%mm3 \n\t" | |
| 144 "packssdw %%mm1, %%mm0 \n\t" | |
| 145 "packssdw %%mm3, %%mm2 \n\t" | |
| 146 "movq %%mm0, 24(%0, %%edi) \n\t" | |
| 147 "movq %%mm2, 32(%0, %%edi) \n\t" | |
| 148 | |
| 149 "addl $16, %%esi \n\t" | |
| 150 " jnz 1b \n\t" | |
| 151 "emms \n\t" | |
| 152 :: "r" (s16+1280), "r" (f+256) | |
| 153 :"%esi", "%edi", "memory" | |
| 154 ); | |
| 3909 | 155 return 5*256; |
| 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; | |
| 3569 | 161 asm volatile( |
| 162 "movl $-1024, %%esi \n\t" | |
| 163 "movq magicF2W, %%mm7 \n\t" | |
| 164 "1: \n\t" | |
| 165 "movq (%1, %%esi), %%mm0 \n\t" | |
| 166 "movq 8(%1, %%esi), %%mm1 \n\t" | |
| 167 "movq 1024(%1, %%esi), %%mm2 \n\t" | |
| 168 "movq 1032(%1, %%esi), %%mm3 \n\t" | |
| 169 "psubd %%mm7, %%mm0 \n\t" | |
| 170 "psubd %%mm7, %%mm1 \n\t" | |
| 171 "psubd %%mm7, %%mm2 \n\t" | |
| 172 "psubd %%mm7, %%mm3 \n\t" | |
| 173 "packssdw %%mm1, %%mm0 \n\t" | |
| 174 "packssdw %%mm3, %%mm2 \n\t" | |
| 175 "movq 2048(%1, %%esi), %%mm3 \n\t" | |
| 176 "movq 2056(%1, %%esi), %%mm4 \n\t" | |
| 177 "movq 3072(%1, %%esi), %%mm5 \n\t" | |
| 178 "movq 3080(%1, %%esi), %%mm6 \n\t" | |
| 179 "psubd %%mm7, %%mm3 \n\t" | |
| 180 "psubd %%mm7, %%mm4 \n\t" | |
| 181 "psubd %%mm7, %%mm5 \n\t" | |
| 182 "psubd %%mm7, %%mm6 \n\t" | |
| 183 "packssdw %%mm4, %%mm3 \n\t" | |
| 184 "packssdw %%mm6, %%mm5 \n\t" | |
| 185 "movq %%mm0, %%mm1 \n\t" | |
| 186 "movq %%mm3, %%mm4 \n\t" | |
| 187 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 188 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 189 "punpcklwd %%mm5, %%mm3 \n\t" | |
| 190 "punpckhwd %%mm5, %%mm4 \n\t" | |
| 191 "movq %%mm0, %%mm2 \n\t" | |
| 192 "movq %%mm1, %%mm5 \n\t" | |
| 193 "punpckldq %%mm3, %%mm0 \n\t" | |
| 194 "punpckhdq %%mm3, %%mm2 \n\t" | |
| 195 "punpckldq %%mm4, %%mm1 \n\t" | |
| 196 "punpckhdq %%mm4, %%mm5 \n\t" | |
| 197 "movq %%mm0, (%0, %%esi,2) \n\t" | |
| 198 "movq %%mm2, 8(%0, %%esi,2) \n\t" | |
| 199 "movq %%mm1, 16(%0, %%esi,2) \n\t" | |
| 200 "movq %%mm5, 24(%0, %%esi,2) \n\t" | |
| 201 "addl $16, %%esi \n\t" | |
| 202 " jnz 1b \n\t" | |
| 203 "emms \n\t" | |
| 204 :: "r" (s16+1024), "r" (f+256) | |
| 205 :"%esi", "memory" | |
| 206 ); | |
| 3909 | 207 return 4*256; |
| 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; | |
| 3653 | 213 asm volatile( |
| 214 "movl $-1024, %%esi \n\t" | |
| 215 "movq magicF2W, %%mm7 \n\t" | |
| 216 "1: \n\t" | |
| 217 "movd (%1, %%esi), %%mm0 \n\t" | |
| 218 "punpckldq 2048(%1, %%esi), %%mm0\n\t" | |
| 219 "movd 3072(%1, %%esi), %%mm1 \n\t" | |
| 220 "punpckldq 4096(%1, %%esi), %%mm1\n\t" | |
| 221 "movd 1024(%1, %%esi), %%mm2 \n\t" | |
| 222 "punpckldq 4(%1, %%esi), %%mm2 \n\t" | |
| 223 "movd 2052(%1, %%esi), %%mm3 \n\t" | |
| 224 "punpckldq 3076(%1, %%esi), %%mm3\n\t" | |
| 225 "movd 4100(%1, %%esi), %%mm4 \n\t" | |
| 226 "punpckldq 1028(%1, %%esi), %%mm4\n\t" | |
| 227 "movd 8(%1, %%esi), %%mm5 \n\t" | |
| 228 "punpckldq 2056(%1, %%esi), %%mm5\n\t" | |
| 229 "leal (%%esi, %%esi, 4), %%edi \n\t" | |
| 230 "sarl $1, %%edi \n\t" | |
| 231 "psubd %%mm7, %%mm0 \n\t" | |
| 232 "psubd %%mm7, %%mm1 \n\t" | |
| 233 "psubd %%mm7, %%mm2 \n\t" | |
| 234 "psubd %%mm7, %%mm3 \n\t" | |
| 235 "psubd %%mm7, %%mm4 \n\t" | |
| 236 "psubd %%mm7, %%mm5 \n\t" | |
| 237 "packssdw %%mm1, %%mm0 \n\t" | |
| 238 "packssdw %%mm3, %%mm2 \n\t" | |
| 239 "packssdw %%mm5, %%mm4 \n\t" | |
| 240 "movq %%mm0, (%0, %%edi) \n\t" | |
| 241 "movq %%mm2, 8(%0, %%edi) \n\t" | |
| 242 "movq %%mm4, 16(%0, %%edi) \n\t" | |
| 243 | |
| 244 "movd 3080(%1, %%esi), %%mm0 \n\t" | |
| 245 "punpckldq 4104(%1, %%esi), %%mm0\n\t" | |
| 246 "movd 1032(%1, %%esi), %%mm1 \n\t" | |
| 247 "punpckldq 12(%1, %%esi), %%mm1\n\t" | |
| 248 "movd 2060(%1, %%esi), %%mm2 \n\t" | |
| 249 "punpckldq 3084(%1, %%esi), %%mm2\n\t" | |
| 250 "movd 4108(%1, %%esi), %%mm3 \n\t" | |
| 251 "punpckldq 1036(%1, %%esi), %%mm3\n\t" | |
| 252 "psubd %%mm7, %%mm0 \n\t" | |
| 253 "psubd %%mm7, %%mm1 \n\t" | |
| 254 "psubd %%mm7, %%mm2 \n\t" | |
| 255 "psubd %%mm7, %%mm3 \n\t" | |
| 256 "packssdw %%mm1, %%mm0 \n\t" | |
| 257 "packssdw %%mm3, %%mm2 \n\t" | |
| 258 "movq %%mm0, 24(%0, %%edi) \n\t" | |
| 259 "movq %%mm2, 32(%0, %%edi) \n\t" | |
| 260 | |
| 261 "addl $16, %%esi \n\t" | |
| 262 " jnz 1b \n\t" | |
| 263 "emms \n\t" | |
| 264 :: "r" (s16+1280), "r" (f+256) | |
| 265 :"%esi", "%edi", "memory" | |
| 266 ); | |
| 3909 | 267 return 5*256; |
| 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; | |
| 3569 | 273 asm volatile( |
| 274 "movl $-1024, %%esi \n\t" | |
| 275 "movq magicF2W, %%mm7 \n\t" | |
| 276 "pxor %%mm6, %%mm6 \n\t" | |
| 277 "1: \n\t" | |
| 278 "movq 1024(%1, %%esi), %%mm0 \n\t" | |
| 279 "movq 1032(%1, %%esi), %%mm1 \n\t" | |
| 280 "movq (%1, %%esi), %%mm2 \n\t" | |
| 281 "movq 8(%1, %%esi), %%mm3 \n\t" | |
| 282 "psubd %%mm7, %%mm0 \n\t" | |
| 283 "psubd %%mm7, %%mm1 \n\t" | |
| 284 "psubd %%mm7, %%mm2 \n\t" | |
| 285 "psubd %%mm7, %%mm3 \n\t" | |
| 286 "packssdw %%mm1, %%mm0 \n\t" | |
| 287 "packssdw %%mm3, %%mm2 \n\t" | |
| 288 "movq %%mm0, %%mm1 \n\t" | |
| 289 "punpcklwd %%mm2, %%mm0 \n\t" | |
| 290 "punpckhwd %%mm2, %%mm1 \n\t" | |
| 291 "leal (%%esi, %%esi, 2), %%edi \n\t" | |
| 292 "movq %%mm6, (%0, %%edi) \n\t" | |
| 293 "movd %%mm0, 8(%0, %%edi) \n\t" | |
| 294 "punpckhdq %%mm0, %%mm0 \n\t" | |
| 295 "movq %%mm6, 12(%0, %%edi) \n\t" | |
| 296 "movd %%mm0, 20(%0, %%edi) \n\t" | |
| 297 "movq %%mm6, 24(%0, %%edi) \n\t" | |
| 298 "movd %%mm1, 32(%0, %%edi) \n\t" | |
| 299 "punpckhdq %%mm1, %%mm1 \n\t" | |
| 300 "movq %%mm6, 36(%0, %%edi) \n\t" | |
| 301 "movd %%mm1, 44(%0, %%edi) \n\t" | |
| 302 "addl $16, %%esi \n\t" | |
| 303 " jnz 1b \n\t" | |
| 304 "emms \n\t" | |
| 305 :: "r" (s16+1536), "r" (f+256) | |
| 306 :"%esi", "%edi", "memory" | |
| 307 ); | |
| 3909 | 308 return 6*256; |
| 309 } | |
| 310 | |
| 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; | |
| 3576 | 314 asm volatile( |
| 315 "movl $-1024, %%esi \n\t" | |
| 316 "movq magicF2W, %%mm7 \n\t" | |
| 317 "pxor %%mm6, %%mm6 \n\t" | |
| 318 "1: \n\t" | |
| 319 "movq 1024(%1, %%esi), %%mm0 \n\t" | |
| 320 "movq 2048(%1, %%esi), %%mm1 \n\t" | |
| 321 "movq (%1, %%esi), %%mm5 \n\t" | |
| 322 "psubd %%mm7, %%mm0 \n\t" | |
| 323 "psubd %%mm7, %%mm1 \n\t" | |
| 324 "psubd %%mm7, %%mm5 \n\t" | |
| 325 "leal (%%esi, %%esi, 2), %%edi \n\t" | |
| 326 | |
| 327 "pxor %%mm4, %%mm4 \n\t" | |
| 328 "packssdw %%mm5, %%mm0 \n\t" // FfAa | |
| 329 "packssdw %%mm4, %%mm1 \n\t" // 00Bb | |
| 330 "punpckhwd %%mm0, %%mm4 \n\t" // F0f0 | |
| 331 "punpcklwd %%mm1, %%mm0 \n\t" // BAba | |
| 332 "movq %%mm0, %%mm1 \n\t" // BAba | |
| 333 "punpckldq %%mm4, %%mm3 \n\t" // f0XX | |
| 334 "punpckldq %%mm6, %%mm0 \n\t" // 00ba | |
| 335 "punpckhdq %%mm1, %%mm3 \n\t" // BAf0 | |
| 336 | |
| 337 "movq %%mm0, (%0, %%edi) \n\t" // 00ba | |
| 338 "punpckhdq %%mm4, %%mm0 \n\t" // F000 | |
| 339 "movq %%mm3, 8(%0, %%edi) \n\t" // BAf0 | |
| 340 "movq %%mm0, 16(%0, %%edi) \n\t" // F000 | |
| 341 "addl $8, %%esi \n\t" | |
| 342 " jnz 1b \n\t" | |
| 343 "emms \n\t" | |
| 344 :: "r" (s16+1536), "r" (f+256) | |
| 345 :"%esi", "%edi", "memory" | |
| 346 ); | |
| 3909 | 347 return 6*256; |
| 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; | |
| 3578 | 353 asm volatile( |
| 354 "movl $-1024, %%esi \n\t" | |
| 355 "movq magicF2W, %%mm7 \n\t" | |
| 356 "pxor %%mm6, %%mm6 \n\t" | |
| 357 "1: \n\t" | |
| 358 "movq 1024(%1, %%esi), %%mm0 \n\t" | |
| 359 "movq 3072(%1, %%esi), %%mm1 \n\t" | |
| 360 "movq 2048(%1, %%esi), %%mm4 \n\t" | |
| 361 "movq (%1, %%esi), %%mm5 \n\t" | |
| 362 "psubd %%mm7, %%mm0 \n\t" | |
| 363 "psubd %%mm7, %%mm1 \n\t" | |
| 364 "psubd %%mm7, %%mm4 \n\t" | |
| 365 "psubd %%mm7, %%mm5 \n\t" | |
| 366 "leal (%%esi, %%esi, 2), %%edi \n\t" | |
| 367 | |
| 368 "packssdw %%mm4, %%mm0 \n\t" // EeAa | |
| 369 "packssdw %%mm5, %%mm1 \n\t" // FfBb | |
| 370 "movq %%mm0, %%mm2 \n\t" // EeAa | |
| 371 "punpcklwd %%mm1, %%mm0 \n\t" // BAba | |
| 372 "punpckhwd %%mm1, %%mm2 \n\t" // FEfe | |
| 373 "movq %%mm0, %%mm1 \n\t" // BAba | |
| 374 "punpckldq %%mm6, %%mm0 \n\t" // 00ba | |
| 375 "punpckhdq %%mm1, %%mm1 \n\t" // BABA | |
| 376 | |
| 377 "movq %%mm0, (%0, %%edi) \n\t" | |
| 378 "punpckhdq %%mm2, %%mm0 \n\t" // FE00 | |
| 379 "punpckldq %%mm1, %%mm2 \n\t" // BAfe | |
| 380 "movq %%mm2, 8(%0, %%edi) \n\t" | |
| 381 "movq %%mm0, 16(%0, %%edi) \n\t" | |
| 382 "addl $8, %%esi \n\t" | |
| 383 " jnz 1b \n\t" | |
| 384 "emms \n\t" | |
| 385 :: "r" (s16+1536), "r" (f+256) | |
| 386 :"%esi", "%edi", "memory" | |
| 387 ); | |
| 3909 | 388 return 6*256; |
| 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; | |
| 3577 | 394 asm volatile( |
| 395 "movl $-1024, %%esi \n\t" | |
| 396 "movq magicF2W, %%mm7 \n\t" | |
| 397 // "pxor %%mm6, %%mm6 \n\t" | |
| 398 "1: \n\t" | |
| 399 "movq 1024(%1, %%esi), %%mm0 \n\t" | |
| 400 "movq 2048(%1, %%esi), %%mm1 \n\t" | |
| 401 "movq 3072(%1, %%esi), %%mm2 \n\t" | |
| 402 "movq 4096(%1, %%esi), %%mm3 \n\t" | |
| 403 "movq (%1, %%esi), %%mm5 \n\t" | |
| 404 "psubd %%mm7, %%mm0 \n\t" | |
| 405 "psubd %%mm7, %%mm1 \n\t" | |
| 406 "psubd %%mm7, %%mm2 \n\t" | |
| 407 "psubd %%mm7, %%mm3 \n\t" | |
| 408 "psubd %%mm7, %%mm5 \n\t" | |
| 409 "leal (%%esi, %%esi, 2), %%edi \n\t" | |
| 410 | |
| 411 "packssdw %%mm2, %%mm0 \n\t" // CcAa | |
| 412 "packssdw %%mm3, %%mm1 \n\t" // DdBb | |
| 413 "packssdw %%mm5, %%mm5 \n\t" // FfFf | |
| 414 "movq %%mm0, %%mm2 \n\t" // CcAa | |
| 415 "punpcklwd %%mm1, %%mm0 \n\t" // BAba | |
| 416 "punpckhwd %%mm1, %%mm2 \n\t" // DCdc | |
| 417 "pxor %%mm4, %%mm4 \n\t" // 0000 | |
| 418 "punpcklwd %%mm5, %%mm4 \n\t" // F0f0 | |
| 419 "movq %%mm0, %%mm1 \n\t" // BAba | |
| 420 "movq %%mm4, %%mm3 \n\t" // F0f0 | |
| 421 "punpckldq %%mm2, %%mm0 \n\t" // dcba | |
| 422 "punpckhdq %%mm1, %%mm1 \n\t" // BABA | |
| 423 "punpckldq %%mm1, %%mm4 \n\t" // BAf0 | |
| 424 "punpckhdq %%mm3, %%mm2 \n\t" // F0DC | |
| 425 | |
| 426 "movq %%mm0, (%0, %%edi) \n\t" | |
| 427 "movq %%mm4, 8(%0, %%edi) \n\t" | |
| 428 "movq %%mm2, 16(%0, %%edi) \n\t" | |
| 429 "addl $8, %%esi \n\t" | |
| 430 " jnz 1b \n\t" | |
| 431 "emms \n\t" | |
| 432 :: "r" (s16+1536), "r" (f+256) | |
| 433 :"%esi", "%edi", "memory" | |
| 434 ); | |
| 3909 | 435 return 6*256; |
| 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; | |
| 3575 | 441 asm volatile( |
| 442 "movl $-1024, %%esi \n\t" | |
| 443 "movq magicF2W, %%mm7 \n\t" | |
| 444 // "pxor %%mm6, %%mm6 \n\t" | |
| 445 "1: \n\t" | |
| 446 "movq 1024(%1, %%esi), %%mm0 \n\t" | |
| 447 "movq 3072(%1, %%esi), %%mm1 \n\t" | |
| 448 "movq 4096(%1, %%esi), %%mm2 \n\t" | |
| 449 "movq 5120(%1, %%esi), %%mm3 \n\t" | |
| 450 "movq 2048(%1, %%esi), %%mm4 \n\t" | |
| 451 "movq (%1, %%esi), %%mm5 \n\t" | |
| 452 "psubd %%mm7, %%mm0 \n\t" | |
| 453 "psubd %%mm7, %%mm1 \n\t" | |
| 454 "psubd %%mm7, %%mm2 \n\t" | |
| 455 "psubd %%mm7, %%mm3 \n\t" | |
| 456 "psubd %%mm7, %%mm4 \n\t" | |
| 457 "psubd %%mm7, %%mm5 \n\t" | |
| 458 "leal (%%esi, %%esi, 2), %%edi \n\t" | |
| 459 | |
| 460 "packssdw %%mm2, %%mm0 \n\t" // CcAa | |
| 461 "packssdw %%mm3, %%mm1 \n\t" // DdBb | |
| 462 "packssdw %%mm4, %%mm4 \n\t" // EeEe | |
| 463 "packssdw %%mm5, %%mm5 \n\t" // FfFf | |
| 464 "movq %%mm0, %%mm2 \n\t" // CcAa | |
| 465 "punpcklwd %%mm1, %%mm0 \n\t" // BAba | |
| 466 "punpckhwd %%mm1, %%mm2 \n\t" // DCdc | |
| 467 "punpcklwd %%mm5, %%mm4 \n\t" // FEfe | |
| 468 "movq %%mm0, %%mm1 \n\t" // BAba | |
| 469 "movq %%mm4, %%mm3 \n\t" // FEfe | |
| 470 "punpckldq %%mm2, %%mm0 \n\t" // dcba | |
| 471 "punpckhdq %%mm1, %%mm1 \n\t" // BABA | |
| 472 "punpckldq %%mm1, %%mm4 \n\t" // BAfe | |
| 473 "punpckhdq %%mm3, %%mm2 \n\t" // FEDC | |
| 474 | |
| 475 "movq %%mm0, (%0, %%edi) \n\t" | |
| 476 "movq %%mm4, 8(%0, %%edi) \n\t" | |
| 477 "movq %%mm2, 16(%0, %%edi) \n\t" | |
| 478 "addl $8, %%esi \n\t" | |
| 479 " jnz 1b \n\t" | |
| 480 "emms \n\t" | |
| 481 :: "r" (s16+1536), "r" (f+256) | |
| 482 :"%esi", "%edi", "memory" | |
| 483 ); | |
| 3909 | 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; | |
|
3412
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
523 break; |
|
21d65a4ae3c9
resample.c added - float->int conversion and channel ordering
arpi
parents:
diff
changeset
|
524 } |
| 3909 | 525 return NULL; |
| 3626 | 526 } |
| 527 | |
| 3909 | 528 |