Parent Directory
| 
Revision Log
Revision 1413 - (view) (download)
| 1 : | edgomez | 1413 | /***************************************************************************** |
| 2 : | * | ||
| 3 : | * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | * - Altivec 8bit<->16bit transfer - | ||
| 5 : | * | ||
| 6 : | * Copyright(C) 2004 Christoph Naegeli <chn@kbw.ch> | ||
| 7 : | * | ||
| 8 : | * This program is free software ; you can redistribute it and/or modify | ||
| 9 : | * it under the terms of the GNU General Public License as published by | ||
| 10 : | * the Free Software Foundation ; either version 2 of the License, or | ||
| 11 : | * (at your option) any later version. | ||
| 12 : | * | ||
| 13 : | * This program is distributed in the hope that it will be useful, | ||
| 14 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 15 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 16 : | * GNU General Public License for more details. | ||
| 17 : | * | ||
| 18 : | * You should have received a copy of the GNU General Public License | ||
| 19 : | * along with this program ; if not, write to the Free Software | ||
| 20 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | * | ||
| 22 : | * $Id: mem_transfer_altivec.c,v 1.1 2004-04-05 20:36:37 edgomez Exp $ | ||
| 23 : | * | ||
| 24 : | ****************************************************************************/ | ||
| 25 : | |||
| 26 : | #ifdef HAVE_ALTIVEC_H | ||
| 27 : | #include <altivec.h> | ||
| 28 : | #endif | ||
| 29 : | |||
| 30 : | #include "../../portab.h" | ||
| 31 : | |||
| 32 : | |||
| 33 : | /* Turn this on if you like debugging the alignment */ | ||
| 34 : | #undef DEBUG | ||
| 35 : | |||
| 36 : | #include <stdio.h> | ||
| 37 : | |||
| 38 : | /* | ||
| 39 : | * This Function assumes dst is 16 byte aligned src is unaligned and stride is | ||
| 40 : | * a multiple of 16. | ||
| 41 : | */ | ||
| 42 : | |||
| 43 : | #define COPY8TO16() \ | ||
| 44 : | s = vec_perm(vec_ld(0, src), vec_ld(16, src), perm); /* load the next 8 bytes */ \ | ||
| 45 : | *dst++ = (vector signed short)vec_mergeh(zerovec, s); /* convert and save */ \ | ||
| 46 : | src += stride | ||
| 47 : | |||
| 48 : | void | ||
| 49 : | transfer_8to16copy_altivec_c(vector signed short *dst, | ||
| 50 : | uint8_t * src, | ||
| 51 : | uint32_t stride) | ||
| 52 : | { | ||
| 53 : | register vector unsigned char perm; | ||
| 54 : | register vector unsigned char s; | ||
| 55 : | register vector unsigned char zerovec; | ||
| 56 : | |||
| 57 : | #ifdef DEBUG | ||
| 58 : | /* check the alignment */ | ||
| 59 : | if(((unsigned long)dst) & 0xf) | ||
| 60 : | fprintf(stderr, "transfer_8to16copy_altivec:incorrect align, dst: %x\n", dst); | ||
| 61 : | if(stride & 0xf) | ||
| 62 : | fprintf(stderr, "transfer_8to16copy_altivec:incorrect align, stride: %u\n", stride); | ||
| 63 : | #endif | ||
| 64 : | |||
| 65 : | /* initialisation */ | ||
| 66 : | perm = vec_lvsl(0, src); | ||
| 67 : | zerovec = vec_splat_u8(0); | ||
| 68 : | |||
| 69 : | /* to the actual copy */ | ||
| 70 : | COPY8TO16(); | ||
| 71 : | COPY8TO16(); | ||
| 72 : | COPY8TO16(); | ||
| 73 : | COPY8TO16(); | ||
| 74 : | |||
| 75 : | COPY8TO16(); | ||
| 76 : | COPY8TO16(); | ||
| 77 : | COPY8TO16(); | ||
| 78 : | COPY8TO16(); | ||
| 79 : | } | ||
| 80 : | |||
| 81 : | |||
| 82 : | /* | ||
| 83 : | * This function assumes dst is 8 byte aligned and stride is a multiple of 8 | ||
| 84 : | * src may be unaligned | ||
| 85 : | */ | ||
| 86 : | |||
| 87 : | #define COPY16TO8() \ | ||
| 88 : | s = vec_perm(src[0], src[1], load_src_perm); \ | ||
| 89 : | packed = vec_packsu(s, vec_splat_s16(0)); \ | ||
| 90 : | mask = vec_perm(mask_stencil, mask_stencil, vec_lvsl(0, dst)); \ | ||
| 91 : | packed = vec_perm(packed, packed, vec_lvsl(0, dst)); \ | ||
| 92 : | packed = vec_sel(packed, vec_ld(0, dst), mask); \ | ||
| 93 : | vec_st(packed, 0, dst); \ | ||
| 94 : | src++; \ | ||
| 95 : | dst += stride | ||
| 96 : | |||
| 97 : | void transfer_16to8copy_altivec_c(uint8_t *dst, | ||
| 98 : | vector signed short *src, | ||
| 99 : | uint32_t stride) | ||
| 100 : | { | ||
| 101 : | register vector signed short s; | ||
| 102 : | register vector unsigned char packed; | ||
| 103 : | register vector unsigned char mask_stencil; | ||
| 104 : | register vector unsigned char mask; | ||
| 105 : | register vector unsigned char load_src_perm; | ||
| 106 : | |||
| 107 : | #ifdef DEBUG | ||
| 108 : | /* if this is on, print alignment errors */ | ||
| 109 : | if(((unsigned long) dst) & 0x7) | ||
| 110 : | fprintf(stderr, "transfer_16to8copy_altivec:incorrect align, dst %x\n", dst); | ||
| 111 : | if(stride & 0x7) | ||
| 112 : | fprintf(stderr, "transfer_16to8copy_altivec:incorrect align, stride %u\n", stride); | ||
| 113 : | #endif | ||
| 114 : | /* Initialisation stuff */ | ||
| 115 : | load_src_perm = vec_lvsl(0, (unsigned char*)src); | ||
| 116 : | mask_stencil = vec_pack(vec_splat_u16(0), vec_splat_u16(-1)); | ||
| 117 : | |||
| 118 : | COPY16TO8(); | ||
| 119 : | COPY16TO8(); | ||
| 120 : | COPY16TO8(); | ||
| 121 : | COPY16TO8(); | ||
| 122 : | |||
| 123 : | COPY16TO8(); | ||
| 124 : | COPY16TO8(); | ||
| 125 : | COPY16TO8(); | ||
| 126 : | COPY16TO8(); | ||
| 127 : | } | ||
| 128 : | |||
| 129 : | |||
| 130 : | |||
| 131 : | /* | ||
| 132 : | * This function assumes dst is 8 byte aligned and src is unaligned. Stride has | ||
| 133 : | * to be a multiple of 8 | ||
| 134 : | */ | ||
| 135 : | |||
| 136 : | #define COPY8TO8() \ | ||
| 137 : | tmp = vec_perm(vec_ld(0, src), vec_ld(16, src), vec_lvsl(0, src)); \ | ||
| 138 : | tmp = vec_sel(vec_perm(tmp, tmp, vec_lvsl(0, dst)), vec_ld(0, dst), vec_perm(mask, mask, vec_lvsl(0, dst))); \ | ||
| 139 : | vec_st(tmp, 0, dst); \ | ||
| 140 : | dst += stride; \ | ||
| 141 : | src += stride | ||
| 142 : | |||
| 143 : | void | ||
| 144 : | transfer8x8_copy_altivec_c( uint8_t * dst, | ||
| 145 : | uint8_t * src, | ||
| 146 : | uint32_t stride) | ||
| 147 : | { | ||
| 148 : | register vector unsigned char tmp; | ||
| 149 : | register vector unsigned char mask; | ||
| 150 : | |||
| 151 : | #ifdef DEBUG | ||
| 152 : | if(((unsigned long)dst) & 0x7) | ||
| 153 : | fprintf("transfer8x8_copy_altivec:incorrect align, dst: %x\n", dst); | ||
| 154 : | if(stride & 0x7) | ||
| 155 : | fprintf("transfer8x8_copy_altivec:incorrect stride, stride: %u\n", stride); | ||
| 156 : | #endif | ||
| 157 : | mask = vec_pack(vec_splat_u16(0), vec_splat_u16(-1)); | ||
| 158 : | |||
| 159 : | COPY8TO8(); | ||
| 160 : | COPY8TO8(); | ||
| 161 : | COPY8TO8(); | ||
| 162 : | COPY8TO8(); | ||
| 163 : | |||
| 164 : | COPY8TO8(); | ||
| 165 : | COPY8TO8(); | ||
| 166 : | COPY8TO8(); | ||
| 167 : | COPY8TO8(); | ||
| 168 : | } | ||
| 169 : | |||
| 170 : | |||
| 171 : | /* | ||
| 172 : | * This function assumes dct is 16 bytes aligned, cur and ref are 8 bytes | ||
| 173 : | * aligned and stride is a multiple of 8 | ||
| 174 : | */ | ||
| 175 : | |||
| 176 : | #define SUB8TO16() \ | ||
| 177 : | c = vec_perm(vec_ld(0, cur), vec_ld(16, cur), vec_lvsl(0, cur)); \ | ||
| 178 : | r = vec_perm(vec_ld(0, ref), vec_ld(16, ref), vec_lvsl(0, ref)); \ | ||
| 179 : | t = vec_sel(vec_perm(r, r, vec_lvsl(0, cur)), vec_ld(0, cur), vec_perm(mask, mask, vec_lvsl(0, cur))); \ | ||
| 180 : | vec_st(t, 0, cur); \ | ||
| 181 : | t = vec_splat_u8(0); \ | ||
| 182 : | cs = (vector signed short)vec_mergeh(t, c); \ | ||
| 183 : | rs = (vector signed short)vec_mergeh(t, r); \ | ||
| 184 : | *dct++ = vec_sub(cs, rs); \ | ||
| 185 : | cur += stride; \ | ||
| 186 : | ref += stride | ||
| 187 : | |||
| 188 : | void | ||
| 189 : | transfer_8to16sub_altivec_c(vector signed short *dct, | ||
| 190 : | uint8_t *cur, | ||
| 191 : | uint8_t *ref, | ||
| 192 : | uint32_t stride) | ||
| 193 : | { | ||
| 194 : | vector unsigned char c; | ||
| 195 : | vector unsigned char r; | ||
| 196 : | vector unsigned char t; | ||
| 197 : | vector unsigned char mask; | ||
| 198 : | vector signed short cs; | ||
| 199 : | vector signed short rs; | ||
| 200 : | |||
| 201 : | #ifdef DEBUG | ||
| 202 : | if(((unsigned long)dct) & 0xf) | ||
| 203 : | fprintf(stderr, "transfer_8to16sub_altivec:incorrect align, dct: %x\n", dct); | ||
| 204 : | if(((unsigned long)cur) & 0x7) | ||
| 205 : | fprintf(stderr, "transfer_8to16sub_altivec:incorrect align, cur: %x\n", cur); | ||
| 206 : | if(((unsigned long)ref) & 0x7) | ||
| 207 : | fprintf(stderr, "transfer_8to16sub_altivec:incorrect align, ref: %x\n", ref); | ||
| 208 : | if(stride & 0x7) | ||
| 209 : | fprintf(stderr, "transfer_8to16sub_altivec:incorrect align, stride: %u\n", stride); | ||
| 210 : | #endif | ||
| 211 : | |||
| 212 : | /* Initialisation */ | ||
| 213 : | mask = vec_pack(vec_splat_u16(0), vec_splat_u16(-1)); | ||
| 214 : | |||
| 215 : | SUB8TO16(); | ||
| 216 : | SUB8TO16(); | ||
| 217 : | SUB8TO16(); | ||
| 218 : | SUB8TO16(); | ||
| 219 : | |||
| 220 : | SUB8TO16(); | ||
| 221 : | SUB8TO16(); | ||
| 222 : | SUB8TO16(); | ||
| 223 : | SUB8TO16(); | ||
| 224 : | } | ||
| 225 : | |||
| 226 : | /* | ||
| 227 : | * This function assumes that dct is 16 bytes aligned, cur and ref is 8 bytes aligned | ||
| 228 : | * and stride is a multiple of 8 | ||
| 229 : | */ | ||
| 230 : | |||
| 231 : | #define SUBRO8TO16() \ | ||
| 232 : | c = vec_perm(vec_ld(0, cur), vec_ld(16, cur), vec_lvsl(0, cur)); \ | ||
| 233 : | r = vec_perm(vec_ld(0, ref), vec_ld(16, ref), vec_lvsl(0, ref)); \ | ||
| 234 : | cs = (vector signed short)vec_mergeh(z, c); \ | ||
| 235 : | rs = (vector signed short)vec_mergeh(z, r); \ | ||
| 236 : | *dct++ = vec_sub(cs, rs); \ | ||
| 237 : | cur += stride; \ | ||
| 238 : | ref += stride | ||
| 239 : | |||
| 240 : | void | ||
| 241 : | transfer_8to16subro_altivec_c(vector signed short *dct, | ||
| 242 : | uint8_t *cur, | ||
| 243 : | uint8_t *ref, | ||
| 244 : | uint32_t stride) | ||
| 245 : | { | ||
| 246 : | register vector unsigned char c; | ||
| 247 : | register vector unsigned char r; | ||
| 248 : | register vector unsigned char z; | ||
| 249 : | register vector signed short cs; | ||
| 250 : | register vector signed short rs; | ||
| 251 : | |||
| 252 : | #ifdef DEBUG | ||
| 253 : | /* if this is on, print alignment errors */ | ||
| 254 : | if(((unsigned long)dct) & 0xf) | ||
| 255 : | fprintf(stderr, "transfer_8to16subro_altivec_c:incorrect align, dct: %x\n", dct); | ||
| 256 : | if(((unsigned long)cur) & 0x7) | ||
| 257 : | fprintf(stderr, "transfer_8to16subro_altivec_c:incorrect align, cur: %x\n", cur); | ||
| 258 : | if(((unsigned long)ref) & 0x7) | ||
| 259 : | fprintf(stderr, "transfer_8to16subro_altivec_c:incorrect align, ref: %x\n", ref); | ||
| 260 : | if(stride & 0x7) | ||
| 261 : | fprintf(stderr, "transfer_8to16subro_altivec_c:incorrect align, stride: %u\n", stride); | ||
| 262 : | #endif | ||
| 263 : | |||
| 264 : | z = vec_splat_u8(0); | ||
| 265 : | |||
| 266 : | SUBRO8TO16(); | ||
| 267 : | SUBRO8TO16(); | ||
| 268 : | SUBRO8TO16(); | ||
| 269 : | SUBRO8TO16(); | ||
| 270 : | |||
| 271 : | SUBRO8TO16(); | ||
| 272 : | SUBRO8TO16(); | ||
| 273 : | SUBRO8TO16(); | ||
| 274 : | SUBRO8TO16(); | ||
| 275 : | } | ||
| 276 : | |||
| 277 : | |||
| 278 : | /* | ||
| 279 : | * This function assumes: | ||
| 280 : | * dct: 16 bytes alignment | ||
| 281 : | * cur: 8 bytes alignment | ||
| 282 : | * ref1: unaligned | ||
| 283 : | * ref2: unaligned | ||
| 284 : | * stride: multiple of 8 | ||
| 285 : | */ | ||
| 286 : | |||
| 287 : | #define SUB28TO16() \ | ||
| 288 : | r1 = vec_perm(vec_ld(0, ref1), vec_ld(16, ref1), vec_lvsl(0, ref1)); \ | ||
| 289 : | r2 = vec_perm(vec_ld(0, ref2), vec_ld(16, ref2), vec_lvsl(0, ref2)); \ | ||
| 290 : | c = vec_perm(vec_ld(0, cur), vec_ld(16, cur), vec_lvsl(0, cur)); \ | ||
| 291 : | r = vec_avg(r1, r2); \ | ||
| 292 : | vec_st(vec_sel(r, vec_ld(0, cur), vec_perm(mask, mask, vec_lvsl(0, cur))), 0, cur); \ | ||
| 293 : | cs = (vector signed short)vec_mergeh(vec_splat_u8(0), c); \ | ||
| 294 : | rs = (vector signed short)vec_mergeh(vec_splat_u8(0), r); \ | ||
| 295 : | *dct++ = vec_sub(cs, rs); \ | ||
| 296 : | cur += stride; \ | ||
| 297 : | ref1 += stride; \ | ||
| 298 : | ref2 += stride | ||
| 299 : | |||
| 300 : | void | ||
| 301 : | transfer_8to16sub2_altivec_c(vector signed short *dct, | ||
| 302 : | uint8_t *cur, | ||
| 303 : | uint8_t *ref1, | ||
| 304 : | uint8_t *ref2, | ||
| 305 : | const uint32_t stride) | ||
| 306 : | { | ||
| 307 : | vector unsigned char r1; | ||
| 308 : | vector unsigned char r2; | ||
| 309 : | vector unsigned char r; | ||
| 310 : | vector unsigned char c; | ||
| 311 : | vector unsigned char mask; | ||
| 312 : | vector signed short cs; | ||
| 313 : | vector signed short rs; | ||
| 314 : | |||
| 315 : | #ifdef DEBUG | ||
| 316 : | /* Dump alignment erros if DEBUG is set */ | ||
| 317 : | if(((unsigned long)dct) & 0xf) | ||
| 318 : | fprintf(stderr, "transfer_8to16sub2_altivec_c:incorrect align, dct: %x\n", dct); | ||
| 319 : | if(((unsigned long)cur) & 0x7) | ||
| 320 : | fprintf(stderr, "transfer_8to16sub2_altivec_c:incorrect align, cur: %x\n", cur); | ||
| 321 : | if(stride & 0x7) | ||
| 322 : | fprintf(stderr, "transfer_8to16sub2_altivec_c:incorrect align, dct: %u\n", stride); | ||
| 323 : | #endif | ||
| 324 : | |||
| 325 : | /* Initialisation */ | ||
| 326 : | mask = vec_pack(vec_splat_u16(0), vec_splat_u16(-1)); | ||
| 327 : | |||
| 328 : | SUB28TO16(); | ||
| 329 : | SUB28TO16(); | ||
| 330 : | SUB28TO16(); | ||
| 331 : | SUB28TO16(); | ||
| 332 : | |||
| 333 : | SUB28TO16(); | ||
| 334 : | SUB28TO16(); | ||
| 335 : | SUB28TO16(); | ||
| 336 : | SUB28TO16(); | ||
| 337 : | } | ||
| 338 : | |||
| 339 : | |||
| 340 : | |||
| 341 : | /* | ||
| 342 : | * This function assumes: | ||
| 343 : | * dst: 8 byte aligned | ||
| 344 : | * src: unaligned | ||
| 345 : | * stride: multiple of 8 | ||
| 346 : | */ | ||
| 347 : | |||
| 348 : | #define ADD16TO8() \ | ||
| 349 : | s = vec_perm(vec_ld(0, src), vec_ld(16, src), vec_lvsl(0, src)); \ | ||
| 350 : | d = vec_perm(vec_ld(0, dst), vec_ld(16, dst), vec_lvsl(0, dst)); \ | ||
| 351 : | ds = (vector signed short)vec_mergeh(vec_splat_u8(0), d); \ | ||
| 352 : | ds = vec_add(ds, s); \ | ||
| 353 : | packed = vec_packsu(ds, vec_splat_s16(0)); \ | ||
| 354 : | mask = vec_pack(vec_splat_u16(0), vec_splat_u16(-1)); \ | ||
| 355 : | mask = vec_perm(mask, mask, vec_lvsl(0, dst)); \ | ||
| 356 : | packed = vec_perm(packed, packed, vec_lvsl(0, dst)); \ | ||
| 357 : | packed = vec_sel(packed, vec_ld(0, dst), mask); \ | ||
| 358 : | vec_st(packed, 0, dst); \ | ||
| 359 : | src += 8; \ | ||
| 360 : | dst += stride | ||
| 361 : | |||
| 362 : | void | ||
| 363 : | transfer_16to8add_altivec_c(uint8_t *dst, | ||
| 364 : | int16_t *src, | ||
| 365 : | uint32_t stride) | ||
| 366 : | { | ||
| 367 : | vector signed short s; | ||
| 368 : | vector signed short ds; | ||
| 369 : | vector unsigned char d; | ||
| 370 : | vector unsigned char packed; | ||
| 371 : | vector unsigned char mask; | ||
| 372 : | |||
| 373 : | #ifdef DEBUG | ||
| 374 : | /* if this is set, dump alignment errors */ | ||
| 375 : | if(((unsigned long)dst) & 0x7) | ||
| 376 : | fprintf(stderr, "transfer_16to8add_altivec_c:incorrect align, dst: %x\n", dst); | ||
| 377 : | if(stride & 0x7) | ||
| 378 : | fprintf(stderr, "transfer_16to8add_altivec_c:incorrect align, dst: %u\n", stride); | ||
| 379 : | #endif | ||
| 380 : | |||
| 381 : | ADD16TO8(); | ||
| 382 : | ADD16TO8(); | ||
| 383 : | ADD16TO8(); | ||
| 384 : | ADD16TO8(); | ||
| 385 : | |||
| 386 : | ADD16TO8(); | ||
| 387 : | ADD16TO8(); | ||
| 388 : | ADD16TO8(); | ||
| 389 : | ADD16TO8(); | ||
| 390 : | } |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |