libavcodec/ppc/idct_altivec.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 /*
  22  * NOTE: This code is based on GPL code from the libmpeg2 project.  The
  23  * author, Michel Lespinasses, has given explicit permission to release
  24  * under LGPL as part of FFmpeg.
  25  */
  26
  27 /*
  28  * FFmpeg integration by Dieter Shirley
  29  *
  30  * This file is a direct copy of the AltiVec IDCT module from the libmpeg2
  31  * project.  I've deleted all of the libmpeg2-specific code, renamed the
  32  * functions and reordered the function parameters.  The only change to the
  33  * IDCT function itself was to factor out the partial transposition, and to
  34  * perform a full transpose at the end of the function.
  35  */
  36
  37
  38 #include <stdlib.h>                                      /* malloc(), free() */
  39 #include <string.h>
  40 #include "config.h"
  41 #if HAVE_ALTIVEC_H
  42 #include <altivec.h>
  43 #endif
  44 #include "libavcodec/dsputil.h"
  45 #include "types_altivec.h"
  46 #include "dsputil_ppc.h"
  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
  84 #define IDCT                                                            \
  85     vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
  86     vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
  87     vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
  88     vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
  89     vec_u16 shift;                                                 \
  90                                                                         \
  91     c4 = vec_splat (constants[0], 0);                                   \
  92     a0 = vec_splat (constants[0], 1);                                   \
  93     a1 = vec_splat (constants[0], 2);                                   \
  94     a2 = vec_splat (constants[0], 3);                                   \
  95     mc4 = vec_splat (constants[0], 4);                                  \
  96     ma2 = vec_splat (constants[0], 5);                                  \
  97     bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3);     \
  98                                                                         \
  99     zero = vec_splat_s16 (0);                                           \
 100     shift = vec_splat_u16 (4);                                          \
 101                                                                         \
 102     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
 103     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
 104     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
 105     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
 106     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
 107     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
 108     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
 109     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
 110                                                                         \
 111     IDCT_HALF                                                           \
 112                                                                         \
 113     vx0 = vec_mergeh (vy0, vy4);                                        \
 114     vx1 = vec_mergel (vy0, vy4);                                        \
 115     vx2 = vec_mergeh (vy1, vy5);                                        \
 116     vx3 = vec_mergel (vy1, vy5);                                        \
 117     vx4 = vec_mergeh (vy2, vy6);                                        \
 118     vx5 = vec_mergel (vy2, vy6);                                        \
 119     vx6 = vec_mergeh (vy3, vy7);                                        \
 120     vx7 = vec_mergel (vy3, vy7);                                        \
 121                                                                         \
 122     vy0 = vec_mergeh (vx0, vx4);                                        \
 123     vy1 = vec_mergel (vx0, vx4);                                        \
 124     vy2 = vec_mergeh (vx1, vx5);                                        \
 125     vy3 = vec_mergel (vx1, vx5);                                        \
 126     vy4 = vec_mergeh (vx2, vx6);                                        \
 127     vy5 = vec_mergel (vx2, vx6);                                        \
 128     vy6 = vec_mergeh (vx3, vx7);                                        \
 129     vy7 = vec_mergel (vx3, vx7);                                        \
 130                                                                         \
 131     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
 132     vx1 = vec_mergel (vy0, vy4);                                        \
 133     vx2 = vec_mergeh (vy1, vy5);                                        \
 134     vx3 = vec_mergel (vy1, vy5);                                        \
 135     vx4 = vec_mergeh (vy2, vy6);                                        \
 136     vx5 = vec_mergel (vy2, vy6);                                        \
 137     vx6 = vec_mergeh (vy3, vy7);                                        \
 138     vx7 = vec_mergel (vy3, vy7);                                        \
 139                                                                         \
 140     IDCT_HALF                                                           \
 141                                                                         \
 142     shift = vec_splat_u16 (6);                                          \
 143     vx0 = vec_sra (vy0, shift);                                         \
 144     vx1 = vec_sra (vy1, shift);                                         \
 145     vx2 = vec_sra (vy2, shift);                                         \
 146     vx3 = vec_sra (vy3, shift);                                         \
 147     vx4 = vec_sra (vy4, shift);                                         \
 148     vx5 = vec_sra (vy5, shift);                                         \
 149     vx6 = vec_sra (vy6, shift);                                         \
 150     vx7 = vec_sra (vy7, shift);
 151
 152
 153 static const vec_s16 constants[5] = {
 154     {23170, 13573,  6518, 21895, -23170, -21895,    32,    31},
 155     {16384, 22725, 21407, 19266,  16384,  19266, 21407, 22725},
 156     {22725, 31521, 29692, 26722,  22725,  26722, 29692, 31521},
 157     {21407, 29692, 27969, 25172,  21407,  25172, 27969, 29692},
 158     {19266, 26722, 25172, 22654,  19266,  22654, 25172, 26722}
 159 };
 160
 161 void idct_put_altivec(uint8_t* dest, int stride, vec_s16* block)
 162 {
 163 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
 164     vec_u8 tmp;
 165
 166 #if CONFIG_POWERPC_PERF
 167 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
 168 #endif
 169     IDCT
 170
 171 #define COPY(dest,src)                                          \
 172     tmp = vec_packsu (src, src);                                \
 173     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);       \
 174     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
 175
 176     COPY (dest, vx0)    dest += stride;
 177     COPY (dest, vx1)    dest += stride;
 178     COPY (dest, vx2)    dest += stride;
 179     COPY (dest, vx3)    dest += stride;
 180     COPY (dest, vx4)    dest += stride;
 181     COPY (dest, vx5)    dest += stride;
 182     COPY (dest, vx6)    dest += stride;
 183     COPY (dest, vx7)
 184
 185 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
 186 }
 187
 188 void idct_add_altivec(uint8_t* dest, int stride, vec_s16* block)
 189 {
 190 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
 191     vec_u8 tmp;
 192     vec_s16 tmp2, tmp3;
 193     vec_u8 perm0;
 194     vec_u8 perm1;
 195     vec_u8 p0, p1, p;
 196
 197 #if CONFIG_POWERPC_PERF
 198 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
 199 #endif
 200
 201     IDCT
 202
 203     p0 = vec_lvsl (0, dest);
 204     p1 = vec_lvsl (stride, dest);
 205     p = vec_splat_u8 (-1);
 206     perm0 = vec_mergeh (p, p0);
 207     perm1 = vec_mergeh (p, p1);
 208
 209 #define ADD(dest,src,perm)                                              \
 210     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
 211     tmp = vec_ld (0, dest);                                             \
 212     tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm);       \
 213     tmp3 = vec_adds (tmp2, src);                                        \
 214     tmp = vec_packsu (tmp3, tmp3);                                      \
 215     vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest);               \
 216     vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
 217
 218     ADD (dest, vx0, perm0)      dest += stride;
 219     ADD (dest, vx1, perm1)      dest += stride;
 220     ADD (dest, vx2, perm0)      dest += stride;
 221     ADD (dest, vx3, perm1)      dest += stride;
 222     ADD (dest, vx4, perm0)      dest += stride;
 223     ADD (dest, vx5, perm1)      dest += stride;
 224     ADD (dest, vx6, perm0)      dest += stride;
 225     ADD (dest, vx7, perm1)
 226
 227 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
 228 }
 229