Parent Directory
| 
Revision Log
Revision 195 - (view) (download)
| 1 : | canard | 98 | /* |
| 2 : | |||
| 3 : | Copyright (C) 2002 Benjamin Herrenschmidt <benh@kernel.crashing.org> | ||
| 4 : | |||
| 5 : | This program is free software; you can redistribute it and/or modify | ||
| 6 : | it under the terms of the GNU General Public License as published by | ||
| 7 : | the Free Software Foundation; either version 2 of the License, or | ||
| 8 : | (at your option) any later version. | ||
| 9 : | |||
| 10 : | This program is distributed in the hope that it will be useful, | ||
| 11 : | but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 12 : | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 13 : | GNU General Public License for more details. | ||
| 14 : | |||
| 15 : | You should have received a copy of the GNU General Public License | ||
| 16 : | along with this program; if not, write to the Free Software | ||
| 17 : | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 18 : | |||
| 19 : | |||
| 20 : | edgomez | 195 | $Id: sad_altivec.c,v 1.3 2002-06-12 20:38:40 edgomez Exp $ |
| 21 : | canard | 98 | $Source: /home/xvid/cvs_copy/cvs-server-root/xvid/xvidcore/src/motion/ppc_asm/sad_altivec.c,v $ |
| 22 : | edgomez | 195 | $Date: 2002-06-12 20:38:40 $ |
| 23 : | $Author: edgomez $ | ||
| 24 : | canard | 98 | |
| 25 : | */ | ||
| 26 : | |||
| 27 : | canard | 115 | #define G_REG |
| 28 : | |||
| 29 : | #ifdef G_REG | ||
| 30 : | edgomez | 195 | register vector unsigned char perm0 asm("%v29"); |
| 31 : | register vector unsigned char perm1 asm("%v30"); | ||
| 32 : | register vector unsigned int zerovec asm("%v31"); | ||
| 33 : | canard | 115 | #endif |
| 34 : | |||
| 35 : | canard | 98 | #include <stdio.h> |
| 36 : | |||
| 37 : | #undef DEBUG | ||
| 38 : | |||
| 39 : | canard | 115 | static const vector unsigned char perms[2] = { |
| 40 : | edgomez | 195 | (vector unsigned char) ( /* Used when cur is aligned */ |
| 41 : | 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, | ||
| 42 : | 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17), | ||
| 43 : | (vector unsigned char) ( /* Used when cur is unaligned */ | ||
| 44 : | 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, | ||
| 45 : | 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f), | ||
| 46 : | canard | 115 | }; |
| 47 : | |||
| 48 : | #ifdef G_REG | ||
| 49 : | edgomez | 195 | void |
| 50 : | sadInit_altivec(void) | ||
| 51 : | canard | 115 | { |
| 52 : | perm0 = perms[0]; | ||
| 53 : | perm1 = perms[1]; | ||
| 54 : | edgomez | 195 | zerovec = (vector unsigned int) (0); |
| 55 : | canard | 115 | } |
| 56 : | edgomez | 195 | static inline const vector unsigned char |
| 57 : | get_perm(unsigned long i) | ||
| 58 : | canard | 115 | { |
| 59 : | return i ? perm1 : perm0; | ||
| 60 : | } | ||
| 61 : | edgomez | 195 | |
| 62 : | canard | 115 | #define ZERODEF |
| 63 : | #define ZEROVEC zerovec | ||
| 64 : | #else | ||
| 65 : | edgomez | 195 | void |
| 66 : | sadInit_altivec(void) | ||
| 67 : | canard | 115 | { |
| 68 : | edgomez | 195 | } |
| 69 : | static inline const vector unsigned char | ||
| 70 : | get_perm(unsigned long i) | ||
| 71 : | { | ||
| 72 : | canard | 115 | return perms[i]; |
| 73 : | } | ||
| 74 : | edgomez | 195 | |
| 75 : | canard | 115 | #define ZERODEF vector unsigned int zerovec = (vector unsigned int)(0) |
| 76 : | #define ZEROVEC zerovec | ||
| 77 : | #endif | ||
| 78 : | |||
| 79 : | |||
| 80 : | canard | 98 | #define SAD16() \ |
| 81 : | t1 = vec_perm(ref[0], ref[1], perm); /* align current vector */ \ | ||
| 82 : | t2 = vec_max(t1, *cur); /* find largest of two */ \ | ||
| 83 : | t3 = vec_min(t1, *cur); /* find smaller of two */ \ | ||
| 84 : | t4 = vec_sub(t2, t3); /* find absolute difference */ \ | ||
| 85 : | sad = vec_sum4s(t4, sad); /* accumulate sum of differences */ \ | ||
| 86 : | cur += stride; ref += stride; | ||
| 87 : | |||
| 88 : | /* | ||
| 89 : | * This function assumes cur and stride are 16 bytes aligned and ref is unaligned | ||
| 90 : | */ | ||
| 91 : | unsigned long | ||
| 92 : | edgomez | 195 | sad16_altivec(const vector unsigned char *cur, |
| 93 : | const vector unsigned char *ref, | ||
| 94 : | unsigned long stride, | ||
| 95 : | const unsigned long best_sad) | ||
| 96 : | canard | 98 | { |
| 97 : | edgomez | 195 | vector unsigned char perm; |
| 98 : | vector unsigned char t1, t2, t3, t4; | ||
| 99 : | vector unsigned int sad; | ||
| 100 : | vector signed int sumdiffs, best_vec; | ||
| 101 : | unsigned long result; | ||
| 102 : | |||
| 103 : | ZERODEF; | ||
| 104 : | |||
| 105 : | canard | 98 | #ifdef DEBUG |
| 106 : | edgomez | 195 | if (((unsigned long) cur) & 0xf) |
| 107 : | fprintf(stderr, "sad16_altivec:incorrect align, cur: %x\n", cur); | ||
| 108 : | canard | 98 | // if (((unsigned long)ref) & 0xf) |
| 109 : | edgomez | 195 | // fprintf(stderr, "sad16_altivec:incorrect align, ref: %x\n", ref); |
| 110 : | if (stride & 0xf) | ||
| 111 : | fprintf(stderr, "sad16_altivec:incorrect align, stride: %x\n", stride); | ||
| 112 : | #endif | ||
| 113 : | /* initialization */ | ||
| 114 : | sad = (vector unsigned int) (ZEROVEC); | ||
| 115 : | stride >>= 4; | ||
| 116 : | perm = vec_lvsl(0, (unsigned char *) ref); | ||
| 117 : | *((unsigned long *) &best_vec) = best_sad; | ||
| 118 : | best_vec = vec_splat(best_vec, 0); | ||
| 119 : | |||
| 120 : | /* perform sum of differences between current and previous */ | ||
| 121 : | SAD16(); | ||
| 122 : | SAD16(); | ||
| 123 : | SAD16(); | ||
| 124 : | SAD16(); | ||
| 125 : | /* Temp sum for exit */ | ||
| 126 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) ZEROVEC); | ||
| 127 : | if (vec_all_ge(sumdiffs, best_vec)) | ||
| 128 : | goto bail; | ||
| 129 : | SAD16(); | ||
| 130 : | SAD16(); | ||
| 131 : | SAD16(); | ||
| 132 : | SAD16(); | ||
| 133 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) ZEROVEC); | ||
| 134 : | if (vec_all_ge(sumdiffs, best_vec)) | ||
| 135 : | goto bail; | ||
| 136 : | SAD16(); | ||
| 137 : | SAD16(); | ||
| 138 : | SAD16(); | ||
| 139 : | SAD16(); | ||
| 140 : | SAD16(); | ||
| 141 : | SAD16(); | ||
| 142 : | SAD16(); | ||
| 143 : | SAD16(); | ||
| 144 : | |||
| 145 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 146 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) ZEROVEC); | ||
| 147 : | bail: | ||
| 148 : | /* copy vector sum into unaligned result */ | ||
| 149 : | sumdiffs = vec_splat(sumdiffs, 3); | ||
| 150 : | vec_ste(sumdiffs, 0, (int *) &result); | ||
| 151 : | return (result); | ||
| 152 : | canard | 98 | } |
| 153 : | |||
| 154 : | #define SAD8() \ | ||
| 155 : | t1 = vec_perm(cur[0], cur[stride], perm_cur); /* align current vector */ \ | ||
| 156 : | t2 = vec_perm(ref[0], ref[1], perm_ref1); /* align current vector */ \ | ||
| 157 : | tp = vec_perm(ref[stride], ref[stride+1], perm_ref1); /* align current vector */ \ | ||
| 158 : | t2 = vec_perm(t2,tp,perm_ref2); \ | ||
| 159 : | t3 = vec_max(t1, t2); /* find largest of two */ \ | ||
| 160 : | t4 = vec_min(t1, t2); /* find smaller of two */ \ | ||
| 161 : | t5 = vec_sub(t3, t4); /* find absolute difference */ \ | ||
| 162 : | sad = vec_sum4s(t5, sad); /* accumulate sum of differences */ \ | ||
| 163 : | cur += stride<<1; ref += stride<<1; | ||
| 164 : | |||
| 165 : | /* | ||
| 166 : | * This function assumes cur is 8 bytes aligned, stride is 16 bytes | ||
| 167 : | * aligned and ref is unaligned | ||
| 168 : | */ | ||
| 169 : | unsigned long | ||
| 170 : | edgomez | 195 | sad8_altivec(const vector unsigned char *cur, |
| 171 : | const vector unsigned char *ref, | ||
| 172 : | unsigned long stride) | ||
| 173 : | canard | 98 | { |
| 174 : | edgomez | 195 | vector unsigned char t1, t2, t3, t4, t5, tp; |
| 175 : | vector unsigned int sad; | ||
| 176 : | vector signed int sumdiffs; | ||
| 177 : | vector unsigned char perm_cur; | ||
| 178 : | vector unsigned char perm_ref1, perm_ref2; | ||
| 179 : | unsigned long result; | ||
| 180 : | canard | 98 | |
| 181 : | edgomez | 195 | ZERODEF; |
| 182 : | |||
| 183 : | canard | 98 | #ifdef DEBUG |
| 184 : | edgomez | 195 | if (((unsigned long) cur) & 0x7) |
| 185 : | fprintf(stderr, "sad8_altivec:incorrect align, cur: %x\n", cur); | ||
| 186 : | canard | 98 | // if (((unsigned long)ref) & 0x7) |
| 187 : | edgomez | 195 | // fprintf(stderr, "sad8_altivec:incorrect align, ref: %x\n", ref); |
| 188 : | if (stride & 0xf) | ||
| 189 : | fprintf(stderr, "sad8_altivec:incorrect align, stride: %x\n", stride); | ||
| 190 : | #endif | ||
| 191 : | canard | 98 | |
| 192 : | edgomez | 195 | perm_cur = get_perm((((unsigned long) cur) >> 3) & 0x01); |
| 193 : | perm_ref1 = vec_lvsl(0, (unsigned char *) ref); | ||
| 194 : | perm_ref2 = get_perm(0); | ||
| 195 : | canard | 98 | |
| 196 : | edgomez | 195 | /* initialization */ |
| 197 : | sad = (vector unsigned int) (ZEROVEC); | ||
| 198 : | stride >>= 4; | ||
| 199 : | |||
| 200 : | /* perform sum of differences between current and previous */ | ||
| 201 : | SAD8(); | ||
| 202 : | SAD8(); | ||
| 203 : | SAD8(); | ||
| 204 : | SAD8(); | ||
| 205 : | |||
| 206 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 207 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) ZEROVEC); | ||
| 208 : | |||
| 209 : | /* copy vector sum into unaligned result */ | ||
| 210 : | sumdiffs = vec_splat(sumdiffs, 3); | ||
| 211 : | vec_ste(sumdiffs, 0, (int *) &result); | ||
| 212 : | return (result); | ||
| 213 : | canard | 98 | } |
| 214 : | |||
| 215 : | #define MEAN16(i)\ | ||
| 216 : | c##i=*cur;\ | ||
| 217 : | mean = vec_sum4s(c##i,mean);\ | ||
| 218 : | cur += stride; | ||
| 219 : | |||
| 220 : | #define DEV16(i) \ | ||
| 221 : | t2 = vec_max(c##i, mn); /* find largest of two */ \ | ||
| 222 : | t3 = vec_min(c##i, mn); /* find smaller of two */ \ | ||
| 223 : | t4 = vec_sub(t2, t3); /* find absolute difference */ \ | ||
| 224 : | dev = vec_sum4s(t4, dev); | ||
| 225 : | |||
| 226 : | unsigned long | ||
| 227 : | edgomez | 195 | dev16_altivec(const vector unsigned char *cur, |
| 228 : | unsigned long stride) | ||
| 229 : | canard | 98 | { |
| 230 : | edgomez | 195 | vector unsigned char t2, t3, t4, mn; |
| 231 : | vector unsigned int mean, dev; | ||
| 232 : | vector signed int sumdiffs; | ||
| 233 : | vector unsigned char c0, c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11, c12, | ||
| 234 : | c13, c14, c15; | ||
| 235 : | unsigned long result; | ||
| 236 : | canard | 98 | |
| 237 : | edgomez | 195 | ZERODEF; |
| 238 : | canard | 98 | |
| 239 : | edgomez | 195 | mean = (vector unsigned int) (ZEROVEC); |
| 240 : | dev = (vector unsigned int) (ZEROVEC); | ||
| 241 : | stride >>= 4; | ||
| 242 : | canard | 98 | |
| 243 : | edgomez | 195 | MEAN16(0); |
| 244 : | MEAN16(1); | ||
| 245 : | MEAN16(2); | ||
| 246 : | MEAN16(3); | ||
| 247 : | MEAN16(4); | ||
| 248 : | MEAN16(5); | ||
| 249 : | MEAN16(6); | ||
| 250 : | MEAN16(7); | ||
| 251 : | MEAN16(8); | ||
| 252 : | MEAN16(9); | ||
| 253 : | MEAN16(10); | ||
| 254 : | MEAN16(11); | ||
| 255 : | MEAN16(12); | ||
| 256 : | MEAN16(13); | ||
| 257 : | MEAN16(14); | ||
| 258 : | MEAN16(15); | ||
| 259 : | canard | 98 | |
| 260 : | edgomez | 195 | sumdiffs = vec_sums((vector signed int) mean, (vector signed int) ZEROVEC); |
| 261 : | mn = vec_perm((vector unsigned char) sumdiffs, | ||
| 262 : | (vector unsigned char) sumdiffs, (vector unsigned char) (14, | ||
| 263 : | 14, | ||
| 264 : | 14, | ||
| 265 : | 14, | ||
| 266 : | 14, | ||
| 267 : | 14, | ||
| 268 : | 14, | ||
| 269 : | 14, | ||
| 270 : | 14, | ||
| 271 : | 14, | ||
| 272 : | 14, | ||
| 273 : | 14, | ||
| 274 : | 14, | ||
| 275 : | 14, | ||
| 276 : | 14, | ||
| 277 : | 14)); | ||
| 278 : | DEV16(0); | ||
| 279 : | DEV16(1); | ||
| 280 : | DEV16(2); | ||
| 281 : | DEV16(3); | ||
| 282 : | DEV16(4); | ||
| 283 : | DEV16(5); | ||
| 284 : | DEV16(6); | ||
| 285 : | DEV16(7); | ||
| 286 : | DEV16(8); | ||
| 287 : | DEV16(9); | ||
| 288 : | DEV16(10); | ||
| 289 : | DEV16(11); | ||
| 290 : | DEV16(12); | ||
| 291 : | DEV16(13); | ||
| 292 : | DEV16(14); | ||
| 293 : | DEV16(15); | ||
| 294 : | |||
| 295 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 296 : | sumdiffs = vec_sums((vector signed int) dev, (vector signed int) ZEROVEC); | ||
| 297 : | |||
| 298 : | /* copy vector sum into unaligned result */ | ||
| 299 : | sumdiffs = vec_splat(sumdiffs, 3); | ||
| 300 : | vec_ste(sumdiffs, 0, (int *) &result); | ||
| 301 : | return (result); | ||
| 302 : | canard | 98 | } |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |