| 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 | /* this code assume that stride % 16 == 0 */ |
| 24 | |
| 25 | #define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) \ |
| 26 | vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc2uc);\ |
| 27 | vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc3uc);\ |
| 28 | \ |
| 29 | psum = vec_mladd(vA, vsrc0ssH, BIAS1);\ |
| 30 | psum = vec_mladd(vB, vsrc1ssH, psum);\ |
| 31 | psum = vec_mladd(vC, vsrc2ssH, psum);\ |
| 32 | psum = vec_mladd(vD, vsrc3ssH, psum);\ |
| 33 | psum = BIAS2(psum);\ |
| 34 | psum = vec_sr(psum, v6us);\ |
| 35 | \ |
| 36 | vdst = vec_ld(0, dst);\ |
| 37 | ppsum = (vec_u8)vec_pack(psum, psum);\ |
| 38 | vfdst = vec_perm(vdst, ppsum, fperm);\ |
| 39 | \ |
| 40 | OP_U8_ALTIVEC(fsum, vfdst, vdst);\ |
| 41 | \ |
| 42 | vec_st(fsum, 0, dst);\ |
| 43 | \ |
| 44 | vsrc0ssH = vsrc2ssH;\ |
| 45 | vsrc1ssH = vsrc3ssH;\ |
| 46 | \ |
| 47 | dst += stride;\ |
| 48 | src += stride; |
| 49 | |
| 50 | #define CHROMA_MC8_ALTIVEC_CORE_SIMPLE \ |
| 51 | \ |
| 52 | vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc);\ |
| 53 | vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc);\ |
| 54 | \ |
| 55 | psum = vec_mladd(vA, vsrc0ssH, v32ss);\ |
| 56 | psum = vec_mladd(vE, vsrc1ssH, psum);\ |
| 57 | psum = vec_sr(psum, v6us);\ |
| 58 | \ |
| 59 | vdst = vec_ld(0, dst);\ |
| 60 | ppsum = (vec_u8)vec_pack(psum, psum);\ |
| 61 | vfdst = vec_perm(vdst, ppsum, fperm);\ |
| 62 | \ |
| 63 | OP_U8_ALTIVEC(fsum, vfdst, vdst);\ |
| 64 | \ |
| 65 | vec_st(fsum, 0, dst);\ |
| 66 | \ |
| 67 | dst += stride;\ |
| 68 | src += stride; |
| 69 | |
| 70 | #define noop(a) a |
| 71 | #define add28(a) vec_add(v28ss, a) |
| 72 | |
| 73 | #ifdef PREFIX_h264_chroma_mc8_altivec |
| 74 | static void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, |
| 75 | int stride, int h, int x, int y) { |
| 76 | DECLARE_ALIGNED(16, signed int, ABCD)[4] = |
| 77 | {((8 - x) * (8 - y)), |
| 78 | (( x) * (8 - y)), |
| 79 | ((8 - x) * ( y)), |
| 80 | (( x) * ( y))}; |
| 81 | register int i; |
| 82 | vec_u8 fperm; |
| 83 | const vec_s32 vABCD = vec_ld(0, ABCD); |
| 84 | const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); |
| 85 | const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); |
| 86 | const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); |
| 87 | const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); |
| 88 | LOAD_ZERO; |
| 89 | const vec_s16 v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5)); |
| 90 | const vec_u16 v6us = vec_splat_u16(6); |
| 91 | register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; |
| 92 | register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; |
| 93 | |
| 94 | vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; |
| 95 | vec_u8 vsrc0uc, vsrc1uc; |
| 96 | vec_s16 vsrc0ssH, vsrc1ssH; |
| 97 | vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; |
| 98 | vec_s16 vsrc2ssH, vsrc3ssH, psum; |
| 99 | vec_u8 vdst, ppsum, vfdst, fsum; |
| 100 | |
| 101 | if (((unsigned long)dst) % 16 == 0) { |
| 102 | fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, |
| 103 | 0x14, 0x15, 0x16, 0x17, |
| 104 | 0x08, 0x09, 0x0A, 0x0B, |
| 105 | 0x0C, 0x0D, 0x0E, 0x0F}; |
| 106 | } else { |
| 107 | fperm = (vec_u8){0x00, 0x01, 0x02, 0x03, |
| 108 | 0x04, 0x05, 0x06, 0x07, |
| 109 | 0x18, 0x19, 0x1A, 0x1B, |
| 110 | 0x1C, 0x1D, 0x1E, 0x1F}; |
| 111 | } |
| 112 | |
| 113 | vsrcAuc = vec_ld(0, src); |
| 114 | |
| 115 | if (loadSecond) |
| 116 | vsrcBuc = vec_ld(16, src); |
| 117 | vsrcperm0 = vec_lvsl(0, src); |
| 118 | vsrcperm1 = vec_lvsl(1, src); |
| 119 | |
| 120 | vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); |
| 121 | if (reallyBadAlign) |
| 122 | vsrc1uc = vsrcBuc; |
| 123 | else |
| 124 | vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); |
| 125 | |
| 126 | vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc0uc); |
| 127 | vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc1uc); |
| 128 | |
| 129 | if (ABCD[3]) { |
| 130 | if (!loadSecond) {// -> !reallyBadAlign |
| 131 | for (i = 0 ; i < h ; i++) { |
| 132 | vsrcCuc = vec_ld(stride + 0, src); |
| 133 | vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); |
| 134 | vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); |
| 135 | |
| 136 | CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) |
| 137 | } |
| 138 | } else { |
| 139 | vec_u8 vsrcDuc; |
| 140 | for (i = 0 ; i < h ; i++) { |
| 141 | vsrcCuc = vec_ld(stride + 0, src); |
| 142 | vsrcDuc = vec_ld(stride + 16, src); |
| 143 | vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); |
| 144 | if (reallyBadAlign) |
| 145 | vsrc3uc = vsrcDuc; |
| 146 | else |
| 147 | vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); |
| 148 | |
| 149 | CHROMA_MC8_ALTIVEC_CORE(v32ss, noop) |
| 150 | } |
| 151 | } |
| 152 | } else { |
| 153 | const vec_s16 vE = vec_add(vB, vC); |
| 154 | if (ABCD[2]) { // x == 0 B == 0 |
| 155 | if (!loadSecond) {// -> !reallyBadAlign |
| 156 | for (i = 0 ; i < h ; i++) { |
| 157 | vsrcCuc = vec_ld(stride + 0, src); |
| 158 | vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); |
| 159 | CHROMA_MC8_ALTIVEC_CORE_SIMPLE |
| 160 | |
| 161 | vsrc0uc = vsrc1uc; |
| 162 | } |
| 163 | } else { |
| 164 | vec_u8 vsrcDuc; |
| 165 | for (i = 0 ; i < h ; i++) { |
| 166 | vsrcCuc = vec_ld(stride + 0, src); |
| 167 | vsrcDuc = vec_ld(stride + 15, src); |
| 168 | vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); |
| 169 | CHROMA_MC8_ALTIVEC_CORE_SIMPLE |
| 170 | |
| 171 | vsrc0uc = vsrc1uc; |
| 172 | } |
| 173 | } |
| 174 | } else { // y == 0 C == 0 |
| 175 | if (!loadSecond) {// -> !reallyBadAlign |
| 176 | for (i = 0 ; i < h ; i++) { |
| 177 | vsrcCuc = vec_ld(0, src); |
| 178 | vsrc0uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); |
| 179 | vsrc1uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); |
| 180 | |
| 181 | CHROMA_MC8_ALTIVEC_CORE_SIMPLE |
| 182 | } |
| 183 | } else { |
| 184 | vec_u8 vsrcDuc; |
| 185 | for (i = 0 ; i < h ; i++) { |
| 186 | vsrcCuc = vec_ld(0, src); |
| 187 | vsrcDuc = vec_ld(15, src); |
| 188 | vsrc0uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); |
| 189 | if (reallyBadAlign) |
| 190 | vsrc1uc = vsrcDuc; |
| 191 | else |
| 192 | vsrc1uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); |
| 193 | |
| 194 | CHROMA_MC8_ALTIVEC_CORE_SIMPLE |
| 195 | } |
| 196 | } |
| 197 | } |
| 198 | } |
| 199 | } |
| 200 | #endif |
| 201 | |
| 202 | /* this code assume that stride % 16 == 0 */ |
| 203 | #ifdef PREFIX_no_rnd_vc1_chroma_mc8_altivec |
| 204 | static void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) { |
| 205 | DECLARE_ALIGNED(16, signed int, ABCD)[4] = |
| 206 | {((8 - x) * (8 - y)), |
| 207 | (( x) * (8 - y)), |
| 208 | ((8 - x) * ( y)), |
| 209 | (( x) * ( y))}; |
| 210 | register int i; |
| 211 | vec_u8 fperm; |
| 212 | const vec_s32 vABCD = vec_ld(0, ABCD); |
| 213 | const vec_s16 vA = vec_splat((vec_s16)vABCD, 1); |
| 214 | const vec_s16 vB = vec_splat((vec_s16)vABCD, 3); |
| 215 | const vec_s16 vC = vec_splat((vec_s16)vABCD, 5); |
| 216 | const vec_s16 vD = vec_splat((vec_s16)vABCD, 7); |
| 217 | LOAD_ZERO; |
| 218 | const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4)); |
| 219 | const vec_u16 v6us = vec_splat_u16(6); |
| 220 | register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1; |
| 221 | register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0; |
| 222 | |
| 223 | vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1; |
| 224 | vec_u8 vsrc0uc, vsrc1uc; |
| 225 | vec_s16 vsrc0ssH, vsrc1ssH; |
| 226 | vec_u8 vsrcCuc, vsrc2uc, vsrc3uc; |
| 227 | vec_s16 vsrc2ssH, vsrc3ssH, psum; |
| 228 | vec_u8 vdst, ppsum, vfdst, fsum; |
| 229 | |
| 230 | if (((unsigned long)dst) % 16 == 0) { |
| 231 | fperm = (vec_u8){0x10, 0x11, 0x12, 0x13, |
| 232 | 0x14, 0x15, 0x16, 0x17, |
| 233 | 0x08, 0x09, 0x0A, 0x0B, |
| 234 | 0x0C, 0x0D, 0x0E, 0x0F}; |
| 235 | } else { |
| 236 | fperm = (vec_u8){0x00, 0x01, 0x02, 0x03, |
| 237 | 0x04, 0x05, 0x06, 0x07, |
| 238 | 0x18, 0x19, 0x1A, 0x1B, |
| 239 | 0x1C, 0x1D, 0x1E, 0x1F}; |
| 240 | } |
| 241 | |
| 242 | vsrcAuc = vec_ld(0, src); |
| 243 | |
| 244 | if (loadSecond) |
| 245 | vsrcBuc = vec_ld(16, src); |
| 246 | vsrcperm0 = vec_lvsl(0, src); |
| 247 | vsrcperm1 = vec_lvsl(1, src); |
| 248 | |
| 249 | vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0); |
| 250 | if (reallyBadAlign) |
| 251 | vsrc1uc = vsrcBuc; |
| 252 | else |
| 253 | vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1); |
| 254 | |
| 255 | vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc); |
| 256 | vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc); |
| 257 | |
| 258 | if (!loadSecond) {// -> !reallyBadAlign |
| 259 | for (i = 0 ; i < h ; i++) { |
| 260 | |
| 261 | |
| 262 | vsrcCuc = vec_ld(stride + 0, src); |
| 263 | |
| 264 | vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0); |
| 265 | vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1); |
| 266 | |
| 267 | CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) |
| 268 | } |
| 269 | } else { |
| 270 | vec_u8 vsrcDuc; |
| 271 | for (i = 0 ; i < h ; i++) { |
| 272 | vsrcCuc = vec_ld(stride + 0, src); |
| 273 | vsrcDuc = vec_ld(stride + 16, src); |
| 274 | |
| 275 | vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0); |
| 276 | if (reallyBadAlign) |
| 277 | vsrc3uc = vsrcDuc; |
| 278 | else |
| 279 | vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1); |
| 280 | |
| 281 | CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28) |
| 282 | } |
| 283 | } |
| 284 | } |
| 285 | #endif |
| 286 | |
| 287 | #undef noop |
| 288 | #undef add28 |
| 289 | #undef CHROMA_MC8_ALTIVEC_CORE |