1/* 2 * Copyright (c) 2001 Michel Lespinasse 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/* NOTE: This code is based on GPL code from the libmpeg2 project. The 22 * author, Michel Lespinasses, has given explicit permission to release 23 * under LGPL as part of FFmpeg. 24 * 25 * FFmpeg integration by Dieter Shirley 26 * 27 * This file is a direct copy of the AltiVec IDCT module from the libmpeg2 28 * project. I've deleted all of the libmpeg2-specific code, renamed the 29 * functions and reordered the function parameters. The only change to the 30 * IDCT function itself was to factor out the partial transposition, and to 31 * perform a full transpose at the end of the function. */ 32 33#include "config.h" 34 35#include <stdlib.h> 36#include <string.h> 37 38#include "libavutil/attributes.h" 39#include "libavutil/cpu.h" 40#include "libavutil/ppc/cpu.h" 41#include "libavutil/ppc/util_altivec.h" 42 43#include "libavcodec/idctdsp.h" 44 45#if HAVE_ALTIVEC 46 47#define IDCT_HALF \ 48 /* 1st stage */ \ 49 t1 = vec_mradds(a1, vx7, vx1); \ 50 t8 = vec_mradds(a1, vx1, vec_subs(zero, vx7)); \ 51 t7 = vec_mradds(a2, vx5, vx3); \ 52 t3 = vec_mradds(ma2, vx3, vx5); \ 53 \ 54 /* 2nd stage */ \ 55 t5 = vec_adds(vx0, vx4); \ 56 t0 = vec_subs(vx0, vx4); \ 57 t2 = vec_mradds(a0, vx6, vx2); \ 58 t4 = vec_mradds(a0, vx2, vec_subs(zero, vx6)); \ 59 t6 = vec_adds(t8, t3); \ 60 t3 = vec_subs(t8, t3); \ 61 t8 = vec_subs(t1, t7); \ 62 t1 = vec_adds(t1, t7); \ 63 \ 64 /* 3rd stage */ \ 65 t7 = vec_adds(t5, t2); \ 66 t2 = vec_subs(t5, t2); \ 67 t5 = vec_adds(t0, t4); \ 68 t0 = vec_subs(t0, t4); \ 69 t4 = vec_subs(t8, t3); \ 70 t3 = vec_adds(t8, t3); \ 71 \ 72 /* 4th stage */ \ 73 vy0 = vec_adds(t7, t1); \ 74 vy7 = vec_subs(t7, t1); \ 75 vy1 = vec_mradds(c4, t3, t5); \ 76 vy6 = vec_mradds(mc4, t3, t5); \ 77 vy2 = vec_mradds(c4, t4, t0); \ 78 vy5 = vec_mradds(mc4, t4, t0); \ 79 vy3 = vec_adds(t2, t6); \ 80 vy4 = vec_subs(t2, t6) 81 82#define IDCT \ 83 vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \ 84 vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \ 85 \ 86 vec_s16 c4 = vec_splat(constants[0], 0); \ 87 vec_s16 a0 = vec_splat(constants[0], 1); \ 88 vec_s16 a1 = vec_splat(constants[0], 2); \ 89 vec_s16 a2 = vec_splat(constants[0], 3); \ 90 vec_s16 mc4 = vec_splat(constants[0], 4); \ 91 vec_s16 ma2 = vec_splat(constants[0], 5); \ 92 vec_s16 bias = (vec_s16) vec_splat((vec_s32) constants[0], 3); \ 93 \ 94 vec_s16 zero = vec_splat_s16(0); \ 95 vec_u16 shift = vec_splat_u16(4); \ 96 \ 97 vec_s16 vx0 = vec_mradds(vec_sl(block[0], shift), constants[1], zero); \ 98 vec_s16 vx1 = vec_mradds(vec_sl(block[1], shift), constants[2], zero); \ 99 vec_s16 vx2 = vec_mradds(vec_sl(block[2], shift), constants[3], zero); \ 100 vec_s16 vx3 = vec_mradds(vec_sl(block[3], shift), constants[4], zero); \ 101 vec_s16 vx4 = vec_mradds(vec_sl(block[4], shift), constants[1], zero); \ 102 vec_s16 vx5 = vec_mradds(vec_sl(block[5], shift), constants[4], zero); \ 103 vec_s16 vx6 = vec_mradds(vec_sl(block[6], shift), constants[3], zero); \ 104 vec_s16 vx7 = vec_mradds(vec_sl(block[7], shift), constants[2], zero); \ 105 \ 106 IDCT_HALF; \ 107 \ 108 vx0 = vec_mergeh(vy0, vy4); \ 109 vx1 = vec_mergel(vy0, vy4); \ 110 vx2 = vec_mergeh(vy1, vy5); \ 111 vx3 = vec_mergel(vy1, vy5); \ 112 vx4 = vec_mergeh(vy2, vy6); \ 113 vx5 = vec_mergel(vy2, vy6); \ 114 vx6 = vec_mergeh(vy3, vy7); \ 115 vx7 = vec_mergel(vy3, vy7); \ 116 \ 117 vy0 = vec_mergeh(vx0, vx4); \ 118 vy1 = vec_mergel(vx0, vx4); \ 119 vy2 = vec_mergeh(vx1, vx5); \ 120 vy3 = vec_mergel(vx1, vx5); \ 121 vy4 = vec_mergeh(vx2, vx6); \ 122 vy5 = vec_mergel(vx2, vx6); \ 123 vy6 = vec_mergeh(vx3, vx7); \ 124 vy7 = vec_mergel(vx3, vx7); \ 125 \ 126 vx0 = vec_adds(vec_mergeh(vy0, vy4), bias); \ 127 vx1 = vec_mergel(vy0, vy4); \ 128 vx2 = vec_mergeh(vy1, vy5); \ 129 vx3 = vec_mergel(vy1, vy5); \ 130 vx4 = vec_mergeh(vy2, vy6); \ 131 vx5 = vec_mergel(vy2, vy6); \ 132 vx6 = vec_mergeh(vy3, vy7); \ 133 vx7 = vec_mergel(vy3, vy7); \ 134 \ 135 IDCT_HALF; \ 136 \ 137 shift = vec_splat_u16(6); \ 138 vx0 = vec_sra(vy0, shift); \ 139 vx1 = vec_sra(vy1, shift); \ 140 vx2 = vec_sra(vy2, shift); \ 141 vx3 = vec_sra(vy3, shift); \ 142 vx4 = vec_sra(vy4, shift); \ 143 vx5 = vec_sra(vy5, shift); \ 144 vx6 = vec_sra(vy6, shift); \ 145 vx7 = vec_sra(vy7, shift) 146 147static const vec_s16 constants[5] = { 148 { 23170, 13573, 6518, 21895, -23170, -21895, 32, 31 }, 149 { 16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725 }, 150 { 22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521 }, 151 { 21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692 }, 152 { 19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722 } 153}; 154 155static void idct_altivec(int16_t *blk) 156{ 157 vec_s16 *block = (vec_s16 *) blk; 158 159 IDCT; 160 161 block[0] = vx0; 162 block[1] = vx1; 163 block[2] = vx2; 164 block[3] = vx3; 165 block[4] = vx4; 166 block[5] = vx5; 167 block[6] = vx6; 168 block[7] = vx7; 169} 170 171static void idct_put_altivec(uint8_t *dest, ptrdiff_t stride, int16_t *blk) 172{ 173 vec_s16 *block = (vec_s16 *) blk; 174 vec_u8 tmp; 175 176 IDCT; 177 178#define COPY(dest, src) \ 179 tmp = vec_packsu(src, src); \ 180 vec_ste((vec_u32) tmp, 0, (unsigned int *) dest); \ 181 vec_ste((vec_u32) tmp, 4, (unsigned int *) dest) 182 183 COPY(dest, vx0); 184 dest += stride; 185 COPY(dest, vx1); 186 dest += stride; 187 COPY(dest, vx2); 188 dest += stride; 189 COPY(dest, vx3); 190 dest += stride; 191 COPY(dest, vx4); 192 dest += stride; 193 COPY(dest, vx5); 194 dest += stride; 195 COPY(dest, vx6); 196 dest += stride; 197 COPY(dest, vx7); 198} 199 200static void idct_add_altivec(uint8_t *dest, ptrdiff_t stride, int16_t *blk) 201{ 202 vec_s16 *block = (vec_s16 *) blk; 203 vec_u8 tmp; 204 vec_s16 tmp2, tmp3; 205 vec_u8 perm0; 206 vec_u8 perm1; 207 vec_u8 p0, p1, p; 208 209 IDCT; 210 211#if HAVE_BIGENDIAN 212 p0 = vec_lvsl(0, dest); 213 p1 = vec_lvsl(stride, dest); 214 p = vec_splat_u8(-1); 215 perm0 = vec_mergeh(p, p0); 216 perm1 = vec_mergeh(p, p1); 217#endif 218 219#if HAVE_BIGENDIAN 220#define GET_TMP2(dest, prm) \ 221 tmp = vec_ld(0, dest); \ 222 tmp2 = (vec_s16) vec_perm(tmp, (vec_u8) zero, prm); 223#else 224#define GET_TMP2(dest, prm) \ 225 tmp = vec_vsx_ld(0, dest); \ 226 tmp2 = (vec_s16) vec_mergeh(tmp, (vec_u8) zero) 227#endif 228 229#define ADD(dest, src, perm) \ 230 GET_TMP2(dest, perm); \ 231 tmp3 = vec_adds(tmp2, src); \ 232 tmp = vec_packsu(tmp3, tmp3); \ 233 vec_ste((vec_u32) tmp, 0, (unsigned int *) dest); \ 234 vec_ste((vec_u32) tmp, 4, (unsigned int *) dest) 235 236 ADD(dest, vx0, perm0); 237 dest += stride; 238 ADD(dest, vx1, perm1); 239 dest += stride; 240 ADD(dest, vx2, perm0); 241 dest += stride; 242 ADD(dest, vx3, perm1); 243 dest += stride; 244 ADD(dest, vx4, perm0); 245 dest += stride; 246 ADD(dest, vx5, perm1); 247 dest += stride; 248 ADD(dest, vx6, perm0); 249 dest += stride; 250 ADD(dest, vx7, perm1); 251} 252 253#endif /* HAVE_ALTIVEC */ 254 255av_cold void ff_idctdsp_init_ppc(IDCTDSPContext *c, AVCodecContext *avctx, 256 unsigned high_bit_depth) 257{ 258#if HAVE_ALTIVEC 259 if (!PPC_ALTIVEC(av_get_cpu_flags())) 260 return; 261 262 if (!high_bit_depth && avctx->lowres == 0) { 263 if ((avctx->idct_algo == FF_IDCT_AUTO && !(avctx->flags & AV_CODEC_FLAG_BITEXACT)) || 264 (avctx->idct_algo == FF_IDCT_ALTIVEC)) { 265 c->idct = idct_altivec; 266 c->idct_add = idct_add_altivec; 267 c->idct_put = idct_put_altivec; 268 c->perm_type = FF_IDCT_PERM_TRANSPOSE; 269 } 270 } 271#endif /* HAVE_ALTIVEC */ 272} 273