idctdsp.c

   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 <stdlib.h>
  34 #include <string.h>
  35 #include "config.h"
  36 #if HAVE_ALTIVEC_H
  37 #include <altivec.h>
  38 #endif
  39
  40 #include "libavutil/attributes.h"
  41 #include "libavutil/cpu.h"
  42 #include "libavutil/ppc/cpu.h"
  43 #include "libavutil/ppc/types_altivec.h"
  44 #include "libavcodec/idctdsp.h"
  45
  46 #if HAVE_ALTIVEC
  47
  48 #define IDCT_HALF                                       \
  49     /* 1st stage */                                     \
  50     t1 = vec_mradds(a1, vx7, vx1);                      \
  51     t8 = vec_mradds(a1, vx1, vec_subs(zero, vx7));      \
  52     t7 = vec_mradds(a2, vx5, vx3);                      \
  53     t3 = vec_mradds(ma2, vx3, vx5);                     \
  54                                                         \
  55     /* 2nd stage */                                     \
  56     t5 = vec_adds(vx0, vx4);                            \
  57     t0 = vec_subs(vx0, vx4);                            \
  58     t2 = vec_mradds(a0, vx6, vx2);                      \
  59     t4 = vec_mradds(a0, vx2, vec_subs(zero, vx6));      \
  60     t6 = vec_adds(t8, t3);                              \
  61     t3 = vec_subs(t8, t3);                              \
  62     t8 = vec_subs(t1, t7);                              \
  63     t1 = vec_adds(t1, t7);                              \
  64                                                         \
  65     /* 3rd stage */                                     \
  66     t7 = vec_adds(t5, t2);                              \
  67     t2 = vec_subs(t5, t2);                              \
  68     t5 = vec_adds(t0, t4);                              \
  69     t0 = vec_subs(t0, t4);                              \
  70     t4 = vec_subs(t8, t3);                              \
  71     t3 = vec_adds(t8, t3);                              \
  72                                                         \
  73     /* 4th stage */                                     \
  74     vy0 = vec_adds(t7, t1);                             \
  75     vy7 = vec_subs(t7, t1);                             \
  76     vy1 = vec_mradds(c4, t3, t5);                       \
  77     vy6 = vec_mradds(mc4, t3, t5);                      \
  78     vy2 = vec_mradds(c4, t4, t0);                       \
  79     vy5 = vec_mradds(mc4, t4, t0);                      \
  80     vy3 = vec_adds(t2, t6);                             \
  81     vy4 = vec_subs(t2, t6)
  82
  83 #define IDCT                                                                \
  84     vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                         \
  85     vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8;                             \
  86                                                                             \
  87     vec_s16 c4   = vec_splat(constants[0], 0);                              \
  88     vec_s16 a0   = vec_splat(constants[0], 1);                              \
  89     vec_s16 a1   = vec_splat(constants[0], 2);                              \
  90     vec_s16 a2   = vec_splat(constants[0], 3);                              \
  91     vec_s16 mc4  = vec_splat(constants[0], 4);                              \
  92     vec_s16 ma2  = vec_splat(constants[0], 5);                              \
  93     vec_s16 bias = (vec_s16) vec_splat((vec_s32) constants[0], 3);          \
  94                                                                             \
  95     vec_s16 zero  = vec_splat_s16(0);                                       \
  96     vec_u16 shift = vec_splat_u16(4);                                       \
  97                                                                             \
  98     vec_s16 vx0 = vec_mradds(vec_sl(block[0], shift), constants[1], zero);  \
  99     vec_s16 vx1 = vec_mradds(vec_sl(block[1], shift), constants[2], zero);  \
 100     vec_s16 vx2 = vec_mradds(vec_sl(block[2], shift), constants[3], zero);  \
 101     vec_s16 vx3 = vec_mradds(vec_sl(block[3], shift), constants[4], zero);  \
 102     vec_s16 vx4 = vec_mradds(vec_sl(block[4], shift), constants[1], zero);  \
 103     vec_s16 vx5 = vec_mradds(vec_sl(block[5], shift), constants[4], zero);  \
 104     vec_s16 vx6 = vec_mradds(vec_sl(block[6], shift), constants[3], zero);  \
 105     vec_s16 vx7 = vec_mradds(vec_sl(block[7], shift), constants[2], zero);  \
 106                                                                             \
 107     IDCT_HALF;                                                              \
 108                                                                             \
 109     vx0 = vec_mergeh(vy0, vy4);                                             \
 110     vx1 = vec_mergel(vy0, vy4);                                             \
 111     vx2 = vec_mergeh(vy1, vy5);                                             \
 112     vx3 = vec_mergel(vy1, vy5);                                             \
 113     vx4 = vec_mergeh(vy2, vy6);                                             \
 114     vx5 = vec_mergel(vy2, vy6);                                             \
 115     vx6 = vec_mergeh(vy3, vy7);                                             \
 116     vx7 = vec_mergel(vy3, vy7);                                             \
 117                                                                             \
 118     vy0 = vec_mergeh(vx0, vx4);                                             \
 119     vy1 = vec_mergel(vx0, vx4);                                             \
 120     vy2 = vec_mergeh(vx1, vx5);                                             \
 121     vy3 = vec_mergel(vx1, vx5);                                             \
 122     vy4 = vec_mergeh(vx2, vx6);                                             \
 123     vy5 = vec_mergel(vx2, vx6);                                             \
 124     vy6 = vec_mergeh(vx3, vx7);                                             \
 125     vy7 = vec_mergel(vx3, vx7);                                             \
 126                                                                             \
 127     vx0 = vec_adds(vec_mergeh(vy0, vy4), bias);                             \
 128     vx1 = vec_mergel(vy0, vy4);                                             \
 129     vx2 = vec_mergeh(vy1, vy5);                                             \
 130     vx3 = vec_mergel(vy1, vy5);                                             \
 131     vx4 = vec_mergeh(vy2, vy6);                                             \
 132     vx5 = vec_mergel(vy2, vy6);                                             \
 133     vx6 = vec_mergeh(vy3, vy7);                                             \
 134     vx7 = vec_mergel(vy3, vy7);                                             \
 135                                                                             \
 136     IDCT_HALF;                                                              \
 137                                                                             \
 138     shift = vec_splat_u16(6);                                               \
 139     vx0 = vec_sra(vy0, shift);                                              \
 140     vx1 = vec_sra(vy1, shift);                                              \
 141     vx2 = vec_sra(vy2, shift);                                              \
 142     vx3 = vec_sra(vy3, shift);                                              \
 143     vx4 = vec_sra(vy4, shift);                                              \
 144     vx5 = vec_sra(vy5, shift);                                              \
 145     vx6 = vec_sra(vy6, shift);                                              \
 146     vx7 = vec_sra(vy7, shift)
 147
 148 static const vec_s16 constants[5] = {
 149     { 23170, 13573,  6518, 21895, -23170, -21895,    32,    31 },
 150     { 16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725 },
 151     { 22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521 },
 152     { 21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692 },
 153     { 19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722 }
 154 };
 155
 156 static void idct_altivec(int16_t *blk)
 157 {
 158     vec_s16 *block = (vec_s16 *) blk;
 159
 160     IDCT;
 161
 162     block[0] = vx0;
 163     block[1] = vx1;
 164     block[2] = vx2;
 165     block[3] = vx3;
 166     block[4] = vx4;
 167     block[5] = vx5;
 168     block[6] = vx6;
 169     block[7] = vx7;
 170 }
 171
 172 static void idct_put_altivec(uint8_t *dest, int stride, int16_t *blk)
 173 {
 174     vec_s16 *block = (vec_s16 *) blk;
 175     vec_u8 tmp;
 176
 177     IDCT;
 178
 179 #define COPY(dest, src)                                     \
 180     tmp = vec_packsu(src, src);                             \
 181     vec_ste((vec_u32) tmp, 0, (unsigned int *) dest);       \
 182     vec_ste((vec_u32) tmp, 4, (unsigned int *) dest)
 183
 184     COPY(dest, vx0);
 185     dest += stride;
 186     COPY(dest, vx1);
 187     dest += stride;
 188     COPY(dest, vx2);
 189     dest += stride;
 190     COPY(dest, vx3);
 191     dest += stride;
 192     COPY(dest, vx4);
 193     dest += stride;
 194     COPY(dest, vx5);
 195     dest += stride;
 196     COPY(dest, vx6);
 197     dest += stride;
 198     COPY(dest, vx7);
 199 }
 200
 201 static void idct_add_altivec(uint8_t *dest, int stride, int16_t *blk)
 202 {
 203     vec_s16 *block = (vec_s16 *) blk;
 204     vec_u8 tmp;
 205     vec_s16 tmp2, tmp3;
 206     vec_u8 perm0;
 207     vec_u8 perm1;
 208     vec_u8 p0, p1, p;
 209
 210     IDCT;
 211
 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
 218 #define ADD(dest, src, perm)                                \
 219     /* *(uint64_t *) &tmp = *(uint64_t *) dest; */          \
 220     tmp  = vec_ld(0, dest);                                 \
 221     tmp2 = (vec_s16) vec_perm(tmp, (vec_u8) zero, perm);    \
 222     tmp3 = vec_adds(tmp2, src);                             \
 223     tmp  = vec_packsu(tmp3, tmp3);                          \
 224     vec_ste((vec_u32) tmp, 0, (unsigned int *) dest);       \
 225     vec_ste((vec_u32) tmp, 4, (unsigned int *) dest)
 226
 227     ADD(dest, vx0, perm0);
 228     dest += stride;
 229     ADD(dest, vx1, perm1);
 230     dest += stride;
 231     ADD(dest, vx2, perm0);
 232     dest += stride;
 233     ADD(dest, vx3, perm1);
 234     dest += stride;
 235     ADD(dest, vx4, perm0);
 236     dest += stride;
 237     ADD(dest, vx5, perm1);
 238     dest += stride;
 239     ADD(dest, vx6, perm0);
 240     dest += stride;
 241     ADD(dest, vx7, perm1);
 242 }
 243
 244 #endif /* HAVE_ALTIVEC */
 245
 246 av_cold void ff_idctdsp_init_ppc(IDCTDSPContext *c, AVCodecContext *avctx,
 247                                  unsigned high_bit_depth)
 248 {
 249 #if HAVE_ALTIVEC
 250     if (!PPC_ALTIVEC(av_get_cpu_flags()))
 251         return;
 252
 253     if (!high_bit_depth && avctx->lowres == 0) {
 254         if ((avctx->idct_algo == FF_IDCT_AUTO && !(avctx->flags & CODEC_FLAG_BITEXACT)) ||
 255             (avctx->idct_algo == FF_IDCT_ALTIVEC)) {
 256             c->idct      = idct_altivec;
 257             c->idct_add  = idct_add_altivec;
 258             c->idct_put  = idct_put_altivec;
 259             c->perm_type = FF_IDCT_PERM_TRANSPOSE;
 260         }
 261     }
 262 #endif /* HAVE_ALTIVEC */
 263 }