Parent Directory
| 
Revision Log
Revision 98 - (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 : | $Id: sad_altivec.c,v 1.1 2002-04-03 14:17:05 canard Exp $ | ||
| 21 : | $Source: /home/xvid/cvs_copy/cvs-server-root/xvid/xvidcore/src/motion/ppc_asm/sad_altivec.c,v $ | ||
| 22 : | $Date: 2002-04-03 14:17:05 $ | ||
| 23 : | $Author: canard $ | ||
| 24 : | |||
| 25 : | */ | ||
| 26 : | |||
| 27 : | #include <stdio.h> | ||
| 28 : | |||
| 29 : | #undef DEBUG | ||
| 30 : | |||
| 31 : | #define SAD16() \ | ||
| 32 : | t1 = vec_perm(ref[0], ref[1], perm); /* align current vector */ \ | ||
| 33 : | t2 = vec_max(t1, *cur); /* find largest of two */ \ | ||
| 34 : | t3 = vec_min(t1, *cur); /* find smaller of two */ \ | ||
| 35 : | t4 = vec_sub(t2, t3); /* find absolute difference */ \ | ||
| 36 : | sad = vec_sum4s(t4, sad); /* accumulate sum of differences */ \ | ||
| 37 : | cur += stride; ref += stride; | ||
| 38 : | |||
| 39 : | /* | ||
| 40 : | * This function assumes cur and stride are 16 bytes aligned and ref is unaligned | ||
| 41 : | */ | ||
| 42 : | unsigned long | ||
| 43 : | sad16_altivec( const vector unsigned char * cur, | ||
| 44 : | const vector unsigned char * ref, | ||
| 45 : | unsigned long stride, | ||
| 46 : | const unsigned long best_sad) | ||
| 47 : | { | ||
| 48 : | vector unsigned char perm; | ||
| 49 : | vector unsigned char t1, t2, t3, t4 ; | ||
| 50 : | vector unsigned int sad, zero; | ||
| 51 : | vector signed int sumdiffs, best_vec; | ||
| 52 : | unsigned long result; | ||
| 53 : | |||
| 54 : | #ifdef DEBUG | ||
| 55 : | if (((unsigned long)cur) & 0xf) | ||
| 56 : | fprintf(stderr, "sad16_altivec:incorrect align, cur: %x\n", cur); | ||
| 57 : | // if (((unsigned long)ref) & 0xf) | ||
| 58 : | // fprintf(stderr, "sad16_altivec:incorrect align, ref: %x\n", ref); | ||
| 59 : | if (stride & 0xf) | ||
| 60 : | fprintf(stderr, "sad16_altivec:incorrect align, stride: %x\n", stride); | ||
| 61 : | #endif | ||
| 62 : | /* initialization */ | ||
| 63 : | zero = (vector unsigned int)(0); | ||
| 64 : | sad = (vector unsigned int)(0); | ||
| 65 : | stride >>= 4; | ||
| 66 : | perm = vec_lvsl(0, (unsigned char *)ref); | ||
| 67 : | *((unsigned long *)&best_vec) = best_sad; | ||
| 68 : | best_vec = vec_splat(best_vec, 0); | ||
| 69 : | |||
| 70 : | /* perform sum of differences between current and previous */ | ||
| 71 : | SAD16(); | ||
| 72 : | SAD16(); | ||
| 73 : | SAD16(); | ||
| 74 : | SAD16(); | ||
| 75 : | /* Temp sum for exit */ | ||
| 76 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); | ||
| 77 : | if (vec_all_ge(sumdiffs, best_vec)) | ||
| 78 : | goto bail; | ||
| 79 : | SAD16(); | ||
| 80 : | SAD16(); | ||
| 81 : | SAD16(); | ||
| 82 : | SAD16(); | ||
| 83 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); | ||
| 84 : | if (vec_all_ge(sumdiffs, best_vec)) | ||
| 85 : | goto bail; | ||
| 86 : | SAD16(); | ||
| 87 : | SAD16(); | ||
| 88 : | SAD16(); | ||
| 89 : | SAD16(); | ||
| 90 : | SAD16(); | ||
| 91 : | SAD16(); | ||
| 92 : | SAD16(); | ||
| 93 : | SAD16(); | ||
| 94 : | |||
| 95 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 96 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); | ||
| 97 : | bail: | ||
| 98 : | /* copy vector sum into unaligned result */ | ||
| 99 : | sumdiffs = vec_splat( sumdiffs, 3 ); | ||
| 100 : | vec_ste( sumdiffs, 0, (int *)&result ); | ||
| 101 : | return( result ); | ||
| 102 : | } | ||
| 103 : | |||
| 104 : | #define SAD8() \ | ||
| 105 : | t1 = vec_perm(cur[0], cur[stride], perm_cur); /* align current vector */ \ | ||
| 106 : | t2 = vec_perm(ref[0], ref[1], perm_ref1); /* align current vector */ \ | ||
| 107 : | tp = vec_perm(ref[stride], ref[stride+1], perm_ref1); /* align current vector */ \ | ||
| 108 : | t2 = vec_perm(t2,tp,perm_ref2); \ | ||
| 109 : | t3 = vec_max(t1, t2); /* find largest of two */ \ | ||
| 110 : | t4 = vec_min(t1, t2); /* find smaller of two */ \ | ||
| 111 : | t5 = vec_sub(t3, t4); /* find absolute difference */ \ | ||
| 112 : | sad = vec_sum4s(t5, sad); /* accumulate sum of differences */ \ | ||
| 113 : | cur += stride<<1; ref += stride<<1; | ||
| 114 : | |||
| 115 : | static const vector unsigned char perms[2] = { | ||
| 116 : | (vector unsigned char)( /* Used when cur is aligned */ | ||
| 117 : | 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, | ||
| 118 : | 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17 | ||
| 119 : | ), | ||
| 120 : | (vector unsigned char)( /* Used when cur is unaligned */ | ||
| 121 : | 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, | ||
| 122 : | 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f | ||
| 123 : | ), | ||
| 124 : | }; | ||
| 125 : | |||
| 126 : | /* | ||
| 127 : | * This function assumes cur is 8 bytes aligned, stride is 16 bytes | ||
| 128 : | * aligned and ref is unaligned | ||
| 129 : | */ | ||
| 130 : | unsigned long | ||
| 131 : | sad8_altivec( const vector unsigned char * cur, | ||
| 132 : | const vector unsigned char * ref, | ||
| 133 : | unsigned long stride) | ||
| 134 : | { | ||
| 135 : | vector unsigned char t1, t2, t3, t4, t5, tp ; | ||
| 136 : | vector unsigned int sad, zero; | ||
| 137 : | vector signed int sumdiffs; | ||
| 138 : | vector unsigned char perm_cur; | ||
| 139 : | vector unsigned char perm_ref1, perm_ref2; | ||
| 140 : | unsigned long result; | ||
| 141 : | |||
| 142 : | #ifdef DEBUG | ||
| 143 : | if (((unsigned long)cur) & 0x7) | ||
| 144 : | fprintf(stderr, "sad8_altivec:incorrect align, cur: %x\n", cur); | ||
| 145 : | // if (((unsigned long)ref) & 0x7) | ||
| 146 : | // fprintf(stderr, "sad8_altivec:incorrect align, ref: %x\n", ref); | ||
| 147 : | if (stride & 0xf) | ||
| 148 : | fprintf(stderr, "sad8_altivec:incorrect align, stride: %x\n", stride); | ||
| 149 : | #endif | ||
| 150 : | |||
| 151 : | perm_cur = perms[(((unsigned long)cur)>>3) & 0x01]; | ||
| 152 : | perm_ref1 = vec_lvsl(0, (unsigned char *)ref); | ||
| 153 : | perm_ref2 = perms[0]; | ||
| 154 : | |||
| 155 : | /* initialization */ | ||
| 156 : | zero = (vector unsigned int)(0); | ||
| 157 : | sad = (vector unsigned int)(0); | ||
| 158 : | stride >>= 4; | ||
| 159 : | |||
| 160 : | /* perform sum of differences between current and previous */ | ||
| 161 : | SAD8(); | ||
| 162 : | SAD8(); | ||
| 163 : | SAD8(); | ||
| 164 : | SAD8(); | ||
| 165 : | |||
| 166 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 167 : | sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); | ||
| 168 : | |||
| 169 : | /* copy vector sum into unaligned result */ | ||
| 170 : | sumdiffs = vec_splat( sumdiffs, 3 ); | ||
| 171 : | vec_ste( sumdiffs, 0, (int *)&result ); | ||
| 172 : | return( result ); | ||
| 173 : | } | ||
| 174 : | |||
| 175 : | #define MEAN16(i)\ | ||
| 176 : | c##i=*cur;\ | ||
| 177 : | mean = vec_sum4s(c##i,mean);\ | ||
| 178 : | cur += stride; | ||
| 179 : | |||
| 180 : | #define DEV16(i) \ | ||
| 181 : | t2 = vec_max(c##i, mn); /* find largest of two */ \ | ||
| 182 : | t3 = vec_min(c##i, mn); /* find smaller of two */ \ | ||
| 183 : | t4 = vec_sub(t2, t3); /* find absolute difference */ \ | ||
| 184 : | dev = vec_sum4s(t4, dev); | ||
| 185 : | |||
| 186 : | unsigned long | ||
| 187 : | dev16_altivec( const vector unsigned char * cur, | ||
| 188 : | unsigned long stride) | ||
| 189 : | { | ||
| 190 : | vector unsigned char t2,t3,t4, mn; | ||
| 191 : | vector unsigned int mean, dev, zero; | ||
| 192 : | vector signed int sumdiffs; | ||
| 193 : | vector unsigned char c0,c1,c2,c3,c4,c5,c6,c7,c8,c9,c10,c11,c12,c13,c14,c15; | ||
| 194 : | unsigned long result; | ||
| 195 : | |||
| 196 : | zero = (vector unsigned int)(0); | ||
| 197 : | mean = (vector unsigned int)(0); | ||
| 198 : | dev = (vector unsigned int)(0); | ||
| 199 : | stride >>= 4; | ||
| 200 : | |||
| 201 : | MEAN16(0); | ||
| 202 : | MEAN16(1); | ||
| 203 : | MEAN16(2); | ||
| 204 : | MEAN16(3); | ||
| 205 : | MEAN16(4); | ||
| 206 : | MEAN16(5); | ||
| 207 : | MEAN16(6); | ||
| 208 : | MEAN16(7); | ||
| 209 : | MEAN16(8); | ||
| 210 : | MEAN16(9); | ||
| 211 : | MEAN16(10); | ||
| 212 : | MEAN16(11); | ||
| 213 : | MEAN16(12); | ||
| 214 : | MEAN16(13); | ||
| 215 : | MEAN16(14); | ||
| 216 : | MEAN16(15); | ||
| 217 : | |||
| 218 : | sumdiffs = vec_sums((vector signed int) mean, (vector signed int) zero); | ||
| 219 : | mn = vec_perm((vector unsigned char)sumdiffs, (vector unsigned char)sumdiffs, | ||
| 220 : | (vector unsigned char)(14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14)); | ||
| 221 : | DEV16(0); | ||
| 222 : | DEV16(1); | ||
| 223 : | DEV16(2); | ||
| 224 : | DEV16(3); | ||
| 225 : | DEV16(4); | ||
| 226 : | DEV16(5); | ||
| 227 : | DEV16(6); | ||
| 228 : | DEV16(7); | ||
| 229 : | DEV16(8); | ||
| 230 : | DEV16(9); | ||
| 231 : | DEV16(10); | ||
| 232 : | DEV16(11); | ||
| 233 : | DEV16(12); | ||
| 234 : | DEV16(13); | ||
| 235 : | DEV16(14); | ||
| 236 : | DEV16(15); | ||
| 237 : | |||
| 238 : | /* sum all parts of difference into one 32 bit quantity */ | ||
| 239 : | sumdiffs = vec_sums((vector signed int) dev, (vector signed int) zero); | ||
| 240 : | |||
| 241 : | /* copy vector sum into unaligned result */ | ||
| 242 : | sumdiffs = vec_splat( sumdiffs, 3 ); | ||
| 243 : | vec_ste( sumdiffs, 0, (int *)&result ); | ||
| 244 : | return( result ); | ||
| 245 : | } |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |