| 1 | /* |
| 2 | * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * FFmpeg is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with FFmpeg; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ |
| 20 | |
| 21 | #include "libavutil/mem.h" |
| 22 | |
| 23 | #ifdef DEBUG |
| 24 | #define ASSERT_ALIGNED(ptr) assert(((unsigned long)ptr&0x0000000F)); |
| 25 | #else |
| 26 | #define ASSERT_ALIGNED(ptr) ; |
| 27 | #endif |
| 28 | |
| 29 | /* this code assume stride % 16 == 0 */ |
| 30 | #ifdef PREFIX_h264_qpel16_h_lowpass_altivec |
| 31 | static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t *dst, |
| 32 | const uint8_t *src, |
| 33 | int dstStride, int srcStride) |
| 34 | { |
| 35 | register int i; |
| 36 | |
| 37 | LOAD_ZERO; |
| 38 | const vec_u8 permM2 = vec_lvsl(-2, src); |
| 39 | const vec_u8 permM1 = vec_lvsl(-1, src); |
| 40 | const vec_u8 permP0 = vec_lvsl(+0, src); |
| 41 | const vec_u8 permP1 = vec_lvsl(+1, src); |
| 42 | const vec_u8 permP2 = vec_lvsl(+2, src); |
| 43 | const vec_u8 permP3 = vec_lvsl(+3, src); |
| 44 | const vec_s16 v5ss = vec_splat_s16(5); |
| 45 | const vec_u16 v5us = vec_splat_u16(5); |
| 46 | const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); |
| 47 | const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); |
| 48 | |
| 49 | vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; |
| 50 | |
| 51 | register int align = ((((unsigned long)src) - 2) % 16); |
| 52 | |
| 53 | vec_s16 srcP0A, srcP0B, srcP1A, srcP1B, |
| 54 | srcP2A, srcP2B, srcP3A, srcP3B, |
| 55 | srcM1A, srcM1B, srcM2A, srcM2B, |
| 56 | sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, |
| 57 | pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, |
| 58 | psumA, psumB, sumA, sumB; |
| 59 | |
| 60 | vec_u8 sum, fsum; |
| 61 | |
| 62 | for (i = 0 ; i < 16 ; i ++) { |
| 63 | vec_u8 srcR1 = vec_ld(-2, src); |
| 64 | vec_u8 srcR2 = vec_ld(14, src); |
| 65 | |
| 66 | switch (align) { |
| 67 | default: { |
| 68 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 69 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 70 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 71 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 72 | srcP2 = vec_perm(srcR1, srcR2, permP2); |
| 73 | srcP3 = vec_perm(srcR1, srcR2, permP3); |
| 74 | } break; |
| 75 | case 11: { |
| 76 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 77 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 78 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 79 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 80 | srcP2 = vec_perm(srcR1, srcR2, permP2); |
| 81 | srcP3 = srcR2; |
| 82 | } break; |
| 83 | case 12: { |
| 84 | vec_u8 srcR3 = vec_ld(30, src); |
| 85 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 86 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 87 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 88 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 89 | srcP2 = srcR2; |
| 90 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 91 | } break; |
| 92 | case 13: { |
| 93 | vec_u8 srcR3 = vec_ld(30, src); |
| 94 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 95 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 96 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 97 | srcP1 = srcR2; |
| 98 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 99 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 100 | } break; |
| 101 | case 14: { |
| 102 | vec_u8 srcR3 = vec_ld(30, src); |
| 103 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 104 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 105 | srcP0 = srcR2; |
| 106 | srcP1 = vec_perm(srcR2, srcR3, permP1); |
| 107 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 108 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 109 | } break; |
| 110 | case 15: { |
| 111 | vec_u8 srcR3 = vec_ld(30, src); |
| 112 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 113 | srcM1 = srcR2; |
| 114 | srcP0 = vec_perm(srcR2, srcR3, permP0); |
| 115 | srcP1 = vec_perm(srcR2, srcR3, permP1); |
| 116 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 117 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 118 | } break; |
| 119 | } |
| 120 | |
| 121 | srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); |
| 122 | srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); |
| 123 | srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); |
| 124 | srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); |
| 125 | |
| 126 | srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); |
| 127 | srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); |
| 128 | srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); |
| 129 | srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); |
| 130 | |
| 131 | srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); |
| 132 | srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); |
| 133 | srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); |
| 134 | srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); |
| 135 | |
| 136 | sum1A = vec_adds(srcP0A, srcP1A); |
| 137 | sum1B = vec_adds(srcP0B, srcP1B); |
| 138 | sum2A = vec_adds(srcM1A, srcP2A); |
| 139 | sum2B = vec_adds(srcM1B, srcP2B); |
| 140 | sum3A = vec_adds(srcM2A, srcP3A); |
| 141 | sum3B = vec_adds(srcM2B, srcP3B); |
| 142 | |
| 143 | pp1A = vec_mladd(sum1A, v20ss, v16ss); |
| 144 | pp1B = vec_mladd(sum1B, v20ss, v16ss); |
| 145 | |
| 146 | pp2A = vec_mladd(sum2A, v5ss, zero_s16v); |
| 147 | pp2B = vec_mladd(sum2B, v5ss, zero_s16v); |
| 148 | |
| 149 | pp3A = vec_add(sum3A, pp1A); |
| 150 | pp3B = vec_add(sum3B, pp1B); |
| 151 | |
| 152 | psumA = vec_sub(pp3A, pp2A); |
| 153 | psumB = vec_sub(pp3B, pp2B); |
| 154 | |
| 155 | sumA = vec_sra(psumA, v5us); |
| 156 | sumB = vec_sra(psumB, v5us); |
| 157 | |
| 158 | sum = vec_packsu(sumA, sumB); |
| 159 | |
| 160 | ASSERT_ALIGNED(dst); |
| 161 | |
| 162 | OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); |
| 163 | |
| 164 | vec_st(fsum, 0, dst); |
| 165 | |
| 166 | src += srcStride; |
| 167 | dst += dstStride; |
| 168 | } |
| 169 | } |
| 170 | #endif |
| 171 | |
| 172 | /* this code assume stride % 16 == 0 */ |
| 173 | #ifdef PREFIX_h264_qpel16_v_lowpass_altivec |
| 174 | static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t *dst, |
| 175 | const uint8_t *src, |
| 176 | int dstStride, int srcStride) |
| 177 | { |
| 178 | register int i; |
| 179 | |
| 180 | LOAD_ZERO; |
| 181 | const vec_u8 perm = vec_lvsl(0, src); |
| 182 | const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); |
| 183 | const vec_u16 v5us = vec_splat_u16(5); |
| 184 | const vec_s16 v5ss = vec_splat_s16(5); |
| 185 | const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); |
| 186 | |
| 187 | const uint8_t *srcbis = src - (srcStride * 2); |
| 188 | |
| 189 | const vec_u8 srcM2a = vec_ld(0, srcbis); |
| 190 | const vec_u8 srcM2b = vec_ld(16, srcbis); |
| 191 | const vec_u8 srcM2 = vec_perm(srcM2a, srcM2b, perm); |
| 192 | //srcbis += srcStride; |
| 193 | const vec_u8 srcM1a = vec_ld(0, srcbis += srcStride); |
| 194 | const vec_u8 srcM1b = vec_ld(16, srcbis); |
| 195 | const vec_u8 srcM1 = vec_perm(srcM1a, srcM1b, perm); |
| 196 | //srcbis += srcStride; |
| 197 | const vec_u8 srcP0a = vec_ld(0, srcbis += srcStride); |
| 198 | const vec_u8 srcP0b = vec_ld(16, srcbis); |
| 199 | const vec_u8 srcP0 = vec_perm(srcP0a, srcP0b, perm); |
| 200 | //srcbis += srcStride; |
| 201 | const vec_u8 srcP1a = vec_ld(0, srcbis += srcStride); |
| 202 | const vec_u8 srcP1b = vec_ld(16, srcbis); |
| 203 | const vec_u8 srcP1 = vec_perm(srcP1a, srcP1b, perm); |
| 204 | //srcbis += srcStride; |
| 205 | const vec_u8 srcP2a = vec_ld(0, srcbis += srcStride); |
| 206 | const vec_u8 srcP2b = vec_ld(16, srcbis); |
| 207 | const vec_u8 srcP2 = vec_perm(srcP2a, srcP2b, perm); |
| 208 | //srcbis += srcStride; |
| 209 | |
| 210 | vec_s16 srcM2ssA = (vec_s16) vec_mergeh(zero_u8v, srcM2); |
| 211 | vec_s16 srcM2ssB = (vec_s16) vec_mergel(zero_u8v, srcM2); |
| 212 | vec_s16 srcM1ssA = (vec_s16) vec_mergeh(zero_u8v, srcM1); |
| 213 | vec_s16 srcM1ssB = (vec_s16) vec_mergel(zero_u8v, srcM1); |
| 214 | vec_s16 srcP0ssA = (vec_s16) vec_mergeh(zero_u8v, srcP0); |
| 215 | vec_s16 srcP0ssB = (vec_s16) vec_mergel(zero_u8v, srcP0); |
| 216 | vec_s16 srcP1ssA = (vec_s16) vec_mergeh(zero_u8v, srcP1); |
| 217 | vec_s16 srcP1ssB = (vec_s16) vec_mergel(zero_u8v, srcP1); |
| 218 | vec_s16 srcP2ssA = (vec_s16) vec_mergeh(zero_u8v, srcP2); |
| 219 | vec_s16 srcP2ssB = (vec_s16) vec_mergel(zero_u8v, srcP2); |
| 220 | |
| 221 | vec_s16 pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, |
| 222 | psumA, psumB, sumA, sumB, |
| 223 | srcP3ssA, srcP3ssB, |
| 224 | sum1A, sum1B, sum2A, sum2B, sum3A, sum3B; |
| 225 | |
| 226 | vec_u8 sum, fsum, srcP3a, srcP3b, srcP3; |
| 227 | |
| 228 | for (i = 0 ; i < 16 ; i++) { |
| 229 | srcP3a = vec_ld(0, srcbis += srcStride); |
| 230 | srcP3b = vec_ld(16, srcbis); |
| 231 | srcP3 = vec_perm(srcP3a, srcP3b, perm); |
| 232 | srcP3ssA = (vec_s16) vec_mergeh(zero_u8v, srcP3); |
| 233 | srcP3ssB = (vec_s16) vec_mergel(zero_u8v, srcP3); |
| 234 | //srcbis += srcStride; |
| 235 | |
| 236 | sum1A = vec_adds(srcP0ssA, srcP1ssA); |
| 237 | sum1B = vec_adds(srcP0ssB, srcP1ssB); |
| 238 | sum2A = vec_adds(srcM1ssA, srcP2ssA); |
| 239 | sum2B = vec_adds(srcM1ssB, srcP2ssB); |
| 240 | sum3A = vec_adds(srcM2ssA, srcP3ssA); |
| 241 | sum3B = vec_adds(srcM2ssB, srcP3ssB); |
| 242 | |
| 243 | srcM2ssA = srcM1ssA; |
| 244 | srcM2ssB = srcM1ssB; |
| 245 | srcM1ssA = srcP0ssA; |
| 246 | srcM1ssB = srcP0ssB; |
| 247 | srcP0ssA = srcP1ssA; |
| 248 | srcP0ssB = srcP1ssB; |
| 249 | srcP1ssA = srcP2ssA; |
| 250 | srcP1ssB = srcP2ssB; |
| 251 | srcP2ssA = srcP3ssA; |
| 252 | srcP2ssB = srcP3ssB; |
| 253 | |
| 254 | pp1A = vec_mladd(sum1A, v20ss, v16ss); |
| 255 | pp1B = vec_mladd(sum1B, v20ss, v16ss); |
| 256 | |
| 257 | pp2A = vec_mladd(sum2A, v5ss, zero_s16v); |
| 258 | pp2B = vec_mladd(sum2B, v5ss, zero_s16v); |
| 259 | |
| 260 | pp3A = vec_add(sum3A, pp1A); |
| 261 | pp3B = vec_add(sum3B, pp1B); |
| 262 | |
| 263 | psumA = vec_sub(pp3A, pp2A); |
| 264 | psumB = vec_sub(pp3B, pp2B); |
| 265 | |
| 266 | sumA = vec_sra(psumA, v5us); |
| 267 | sumB = vec_sra(psumB, v5us); |
| 268 | |
| 269 | sum = vec_packsu(sumA, sumB); |
| 270 | |
| 271 | ASSERT_ALIGNED(dst); |
| 272 | |
| 273 | OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); |
| 274 | |
| 275 | vec_st(fsum, 0, dst); |
| 276 | |
| 277 | dst += dstStride; |
| 278 | } |
| 279 | } |
| 280 | #endif |
| 281 | |
| 282 | /* this code assume stride % 16 == 0 *and* tmp is properly aligned */ |
| 283 | #ifdef PREFIX_h264_qpel16_hv_lowpass_altivec |
| 284 | static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t *dst, int16_t *tmp, |
| 285 | const uint8_t *src, |
| 286 | int dstStride, int tmpStride, |
| 287 | int srcStride) |
| 288 | { |
| 289 | register int i; |
| 290 | LOAD_ZERO; |
| 291 | const vec_u8 permM2 = vec_lvsl(-2, src); |
| 292 | const vec_u8 permM1 = vec_lvsl(-1, src); |
| 293 | const vec_u8 permP0 = vec_lvsl(+0, src); |
| 294 | const vec_u8 permP1 = vec_lvsl(+1, src); |
| 295 | const vec_u8 permP2 = vec_lvsl(+2, src); |
| 296 | const vec_u8 permP3 = vec_lvsl(+3, src); |
| 297 | const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); |
| 298 | const vec_u32 v10ui = vec_splat_u32(10); |
| 299 | const vec_s16 v5ss = vec_splat_s16(5); |
| 300 | const vec_s16 v1ss = vec_splat_s16(1); |
| 301 | const vec_s32 v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9)); |
| 302 | const vec_u32 v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4)); |
| 303 | |
| 304 | register int align = ((((unsigned long)src) - 2) % 16); |
| 305 | |
| 306 | vec_s16 srcP0A, srcP0B, srcP1A, srcP1B, |
| 307 | srcP2A, srcP2B, srcP3A, srcP3B, |
| 308 | srcM1A, srcM1B, srcM2A, srcM2B, |
| 309 | sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, |
| 310 | pp1A, pp1B, pp2A, pp2B, psumA, psumB; |
| 311 | |
| 312 | const vec_u8 mperm = (const vec_u8) |
| 313 | {0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, |
| 314 | 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F}; |
| 315 | int16_t *tmpbis = tmp; |
| 316 | |
| 317 | vec_s16 tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB, |
| 318 | tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB, |
| 319 | tmpP2ssA, tmpP2ssB; |
| 320 | |
| 321 | vec_s32 pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo, |
| 322 | pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo, |
| 323 | pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo, |
| 324 | ssumAe, ssumAo, ssumBe, ssumBo; |
| 325 | vec_u8 fsum, sumv, sum; |
| 326 | vec_s16 ssume, ssumo; |
| 327 | |
| 328 | src -= (2 * srcStride); |
| 329 | for (i = 0 ; i < 21 ; i ++) { |
| 330 | vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; |
| 331 | vec_u8 srcR1 = vec_ld(-2, src); |
| 332 | vec_u8 srcR2 = vec_ld(14, src); |
| 333 | |
| 334 | switch (align) { |
| 335 | default: { |
| 336 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 337 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 338 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 339 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 340 | srcP2 = vec_perm(srcR1, srcR2, permP2); |
| 341 | srcP3 = vec_perm(srcR1, srcR2, permP3); |
| 342 | } break; |
| 343 | case 11: { |
| 344 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 345 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 346 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 347 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 348 | srcP2 = vec_perm(srcR1, srcR2, permP2); |
| 349 | srcP3 = srcR2; |
| 350 | } break; |
| 351 | case 12: { |
| 352 | vec_u8 srcR3 = vec_ld(30, src); |
| 353 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 354 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 355 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 356 | srcP1 = vec_perm(srcR1, srcR2, permP1); |
| 357 | srcP2 = srcR2; |
| 358 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 359 | } break; |
| 360 | case 13: { |
| 361 | vec_u8 srcR3 = vec_ld(30, src); |
| 362 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 363 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 364 | srcP0 = vec_perm(srcR1, srcR2, permP0); |
| 365 | srcP1 = srcR2; |
| 366 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 367 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 368 | } break; |
| 369 | case 14: { |
| 370 | vec_u8 srcR3 = vec_ld(30, src); |
| 371 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 372 | srcM1 = vec_perm(srcR1, srcR2, permM1); |
| 373 | srcP0 = srcR2; |
| 374 | srcP1 = vec_perm(srcR2, srcR3, permP1); |
| 375 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 376 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 377 | } break; |
| 378 | case 15: { |
| 379 | vec_u8 srcR3 = vec_ld(30, src); |
| 380 | srcM2 = vec_perm(srcR1, srcR2, permM2); |
| 381 | srcM1 = srcR2; |
| 382 | srcP0 = vec_perm(srcR2, srcR3, permP0); |
| 383 | srcP1 = vec_perm(srcR2, srcR3, permP1); |
| 384 | srcP2 = vec_perm(srcR2, srcR3, permP2); |
| 385 | srcP3 = vec_perm(srcR2, srcR3, permP3); |
| 386 | } break; |
| 387 | } |
| 388 | |
| 389 | srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); |
| 390 | srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); |
| 391 | srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); |
| 392 | srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); |
| 393 | |
| 394 | srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); |
| 395 | srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); |
| 396 | srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); |
| 397 | srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); |
| 398 | |
| 399 | srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); |
| 400 | srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); |
| 401 | srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); |
| 402 | srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); |
| 403 | |
| 404 | sum1A = vec_adds(srcP0A, srcP1A); |
| 405 | sum1B = vec_adds(srcP0B, srcP1B); |
| 406 | sum2A = vec_adds(srcM1A, srcP2A); |
| 407 | sum2B = vec_adds(srcM1B, srcP2B); |
| 408 | sum3A = vec_adds(srcM2A, srcP3A); |
| 409 | sum3B = vec_adds(srcM2B, srcP3B); |
| 410 | |
| 411 | pp1A = vec_mladd(sum1A, v20ss, sum3A); |
| 412 | pp1B = vec_mladd(sum1B, v20ss, sum3B); |
| 413 | |
| 414 | pp2A = vec_mladd(sum2A, v5ss, zero_s16v); |
| 415 | pp2B = vec_mladd(sum2B, v5ss, zero_s16v); |
| 416 | |
| 417 | psumA = vec_sub(pp1A, pp2A); |
| 418 | psumB = vec_sub(pp1B, pp2B); |
| 419 | |
| 420 | vec_st(psumA, 0, tmp); |
| 421 | vec_st(psumB, 16, tmp); |
| 422 | |
| 423 | src += srcStride; |
| 424 | tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ |
| 425 | } |
| 426 | |
| 427 | tmpM2ssA = vec_ld(0, tmpbis); |
| 428 | tmpM2ssB = vec_ld(16, tmpbis); |
| 429 | tmpbis += tmpStride; |
| 430 | tmpM1ssA = vec_ld(0, tmpbis); |
| 431 | tmpM1ssB = vec_ld(16, tmpbis); |
| 432 | tmpbis += tmpStride; |
| 433 | tmpP0ssA = vec_ld(0, tmpbis); |
| 434 | tmpP0ssB = vec_ld(16, tmpbis); |
| 435 | tmpbis += tmpStride; |
| 436 | tmpP1ssA = vec_ld(0, tmpbis); |
| 437 | tmpP1ssB = vec_ld(16, tmpbis); |
| 438 | tmpbis += tmpStride; |
| 439 | tmpP2ssA = vec_ld(0, tmpbis); |
| 440 | tmpP2ssB = vec_ld(16, tmpbis); |
| 441 | tmpbis += tmpStride; |
| 442 | |
| 443 | for (i = 0 ; i < 16 ; i++) { |
| 444 | const vec_s16 tmpP3ssA = vec_ld(0, tmpbis); |
| 445 | const vec_s16 tmpP3ssB = vec_ld(16, tmpbis); |
| 446 | |
| 447 | const vec_s16 sum1A = vec_adds(tmpP0ssA, tmpP1ssA); |
| 448 | const vec_s16 sum1B = vec_adds(tmpP0ssB, tmpP1ssB); |
| 449 | const vec_s16 sum2A = vec_adds(tmpM1ssA, tmpP2ssA); |
| 450 | const vec_s16 sum2B = vec_adds(tmpM1ssB, tmpP2ssB); |
| 451 | const vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA); |
| 452 | const vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB); |
| 453 | |
| 454 | tmpbis += tmpStride; |
| 455 | |
| 456 | tmpM2ssA = tmpM1ssA; |
| 457 | tmpM2ssB = tmpM1ssB; |
| 458 | tmpM1ssA = tmpP0ssA; |
| 459 | tmpM1ssB = tmpP0ssB; |
| 460 | tmpP0ssA = tmpP1ssA; |
| 461 | tmpP0ssB = tmpP1ssB; |
| 462 | tmpP1ssA = tmpP2ssA; |
| 463 | tmpP1ssB = tmpP2ssB; |
| 464 | tmpP2ssA = tmpP3ssA; |
| 465 | tmpP2ssB = tmpP3ssB; |
| 466 | |
| 467 | pp1Ae = vec_mule(sum1A, v20ss); |
| 468 | pp1Ao = vec_mulo(sum1A, v20ss); |
| 469 | pp1Be = vec_mule(sum1B, v20ss); |
| 470 | pp1Bo = vec_mulo(sum1B, v20ss); |
| 471 | |
| 472 | pp2Ae = vec_mule(sum2A, v5ss); |
| 473 | pp2Ao = vec_mulo(sum2A, v5ss); |
| 474 | pp2Be = vec_mule(sum2B, v5ss); |
| 475 | pp2Bo = vec_mulo(sum2B, v5ss); |
| 476 | |
| 477 | pp3Ae = vec_sra((vec_s32)sum3A, v16ui); |
| 478 | pp3Ao = vec_mulo(sum3A, v1ss); |
| 479 | pp3Be = vec_sra((vec_s32)sum3B, v16ui); |
| 480 | pp3Bo = vec_mulo(sum3B, v1ss); |
| 481 | |
| 482 | pp1cAe = vec_add(pp1Ae, v512si); |
| 483 | pp1cAo = vec_add(pp1Ao, v512si); |
| 484 | pp1cBe = vec_add(pp1Be, v512si); |
| 485 | pp1cBo = vec_add(pp1Bo, v512si); |
| 486 | |
| 487 | pp32Ae = vec_sub(pp3Ae, pp2Ae); |
| 488 | pp32Ao = vec_sub(pp3Ao, pp2Ao); |
| 489 | pp32Be = vec_sub(pp3Be, pp2Be); |
| 490 | pp32Bo = vec_sub(pp3Bo, pp2Bo); |
| 491 | |
| 492 | sumAe = vec_add(pp1cAe, pp32Ae); |
| 493 | sumAo = vec_add(pp1cAo, pp32Ao); |
| 494 | sumBe = vec_add(pp1cBe, pp32Be); |
| 495 | sumBo = vec_add(pp1cBo, pp32Bo); |
| 496 | |
| 497 | ssumAe = vec_sra(sumAe, v10ui); |
| 498 | ssumAo = vec_sra(sumAo, v10ui); |
| 499 | ssumBe = vec_sra(sumBe, v10ui); |
| 500 | ssumBo = vec_sra(sumBo, v10ui); |
| 501 | |
| 502 | ssume = vec_packs(ssumAe, ssumBe); |
| 503 | ssumo = vec_packs(ssumAo, ssumBo); |
| 504 | |
| 505 | sumv = vec_packsu(ssume, ssumo); |
| 506 | sum = vec_perm(sumv, sumv, mperm); |
| 507 | |
| 508 | ASSERT_ALIGNED(dst); |
| 509 | |
| 510 | OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); |
| 511 | |
| 512 | vec_st(fsum, 0, dst); |
| 513 | |
| 514 | dst += dstStride; |
| 515 | } |
| 516 | } |
| 517 | #endif |