libavcodec/idctdsp.c

   1 /*
   2  * This file is part of FFmpeg.
   3  *
   4  * FFmpeg is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU Lesser General Public
   6  * License as published by the Free Software Foundation; either
   7  * version 2.1 of the License, or (at your option) any later version.
   8  *
   9  * FFmpeg is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * Lesser General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Lesser General Public
  15  * License along with FFmpeg; if not, write to the Free Software
  16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  17  */
  18
  19 #include "config.h"
  20 #include "config_components.h"
  21 #include "libavutil/attributes.h"
  22 #include "libavutil/common.h"
  23 #include "avcodec.h"
  24 #include "dct.h"
  25 #include "faanidct.h"
  26 #include "idctdsp.h"
  27 #include "simple_idct.h"
  28 #include "xvididct.h"
  29
  30 av_cold void ff_permute_scantable(uint8_t dst[64], const uint8_t src[64],
  31                                   const uint8_t permutation[64])
  32 {
  33     for (int i = 0; i < 64; i++) {
  34         int j = src[i];
  35         dst[i] = permutation[j];
  36     }
  37 }
  38
  39 av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,
  40                                            enum idct_permutation_type perm_type)
  41 {
  42     int i;
  43
  44 #if ARCH_X86
  45     if (ff_init_scantable_permutation_x86(idct_permutation,
  46                                           perm_type))
  47         return;
  48 #endif
  49
  50     switch (perm_type) {
  51     case FF_IDCT_PERM_NONE:
  52         for (i = 0; i < 64; i++)
  53             idct_permutation[i] = i;
  54         break;
  55     case FF_IDCT_PERM_LIBMPEG2:
  56         for (i = 0; i < 64; i++)
  57             idct_permutation[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
  58         break;
  59     case FF_IDCT_PERM_TRANSPOSE:
  60         for (i = 0; i < 64; i++)
  61             idct_permutation[i] = ((i & 7) << 3) | (i >> 3);
  62         break;
  63     case FF_IDCT_PERM_PARTTRANS:
  64         for (i = 0; i < 64; i++)
  65             idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3);
  66         break;
  67     default:
  68         av_log(NULL, AV_LOG_ERROR,
  69                "Internal error, IDCT permutation not set\n");
  70     }
  71 }
  72
  73 void ff_put_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,
  74                              ptrdiff_t line_size)
  75 {
  76     int i;
  77
  78     /* read the pixels */
  79     for (i = 0; i < 8; i++) {
  80         pixels[0] = av_clip_uint8(block[0]);
  81         pixels[1] = av_clip_uint8(block[1]);
  82         pixels[2] = av_clip_uint8(block[2]);
  83         pixels[3] = av_clip_uint8(block[3]);
  84         pixels[4] = av_clip_uint8(block[4]);
  85         pixels[5] = av_clip_uint8(block[5]);
  86         pixels[6] = av_clip_uint8(block[6]);
  87         pixels[7] = av_clip_uint8(block[7]);
  88
  89         pixels += line_size;
  90         block  += 8;
  91     }
  92 }
  93
  94 static void put_pixels_clamped4_c(const int16_t *block, uint8_t *restrict pixels,
  95                                  int line_size)
  96 {
  97     int i;
  98
  99     /* read the pixels */
 100     for(i=0;i<4;i++) {
 101         pixels[0] = av_clip_uint8(block[0]);
 102         pixels[1] = av_clip_uint8(block[1]);
 103         pixels[2] = av_clip_uint8(block[2]);
 104         pixels[3] = av_clip_uint8(block[3]);
 105
 106         pixels += line_size;
 107         block += 8;
 108     }
 109 }
 110
 111 static void put_pixels_clamped2_c(const int16_t *block, uint8_t *restrict pixels,
 112                                  int line_size)
 113 {
 114     int i;
 115
 116     /* read the pixels */
 117     for(i=0;i<2;i++) {
 118         pixels[0] = av_clip_uint8(block[0]);
 119         pixels[1] = av_clip_uint8(block[1]);
 120
 121         pixels += line_size;
 122         block += 8;
 123     }
 124 }
 125
 126 static void put_signed_pixels_clamped_c(const int16_t *block,
 127                                         uint8_t *restrict pixels,
 128                                         ptrdiff_t line_size)
 129 {
 130     int i, j;
 131
 132     for (i = 0; i < 8; i++) {
 133         for (j = 0; j < 8; j++) {
 134             if (*block < -128)
 135                 *pixels = 0;
 136             else if (*block > 127)
 137                 *pixels = 255;
 138             else
 139                 *pixels = (uint8_t) (*block + 128);
 140             block++;
 141             pixels++;
 142         }
 143         pixels += (line_size - 8);
 144     }
 145 }
 146
 147 void ff_add_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,
 148                              ptrdiff_t line_size)
 149 {
 150     int i;
 151
 152     /* read the pixels */
 153     for (i = 0; i < 8; i++) {
 154         pixels[0] = av_clip_uint8(pixels[0] + block[0]);
 155         pixels[1] = av_clip_uint8(pixels[1] + block[1]);
 156         pixels[2] = av_clip_uint8(pixels[2] + block[2]);
 157         pixels[3] = av_clip_uint8(pixels[3] + block[3]);
 158         pixels[4] = av_clip_uint8(pixels[4] + block[4]);
 159         pixels[5] = av_clip_uint8(pixels[5] + block[5]);
 160         pixels[6] = av_clip_uint8(pixels[6] + block[6]);
 161         pixels[7] = av_clip_uint8(pixels[7] + block[7]);
 162         pixels   += line_size;
 163         block    += 8;
 164     }
 165 }
 166
 167 static void add_pixels_clamped4_c(const int16_t *block, uint8_t *restrict pixels,
 168                           int line_size)
 169 {
 170     int i;
 171
 172     /* read the pixels */
 173     for(i=0;i<4;i++) {
 174         pixels[0] = av_clip_uint8(pixels[0] + block[0]);
 175         pixels[1] = av_clip_uint8(pixels[1] + block[1]);
 176         pixels[2] = av_clip_uint8(pixels[2] + block[2]);
 177         pixels[3] = av_clip_uint8(pixels[3] + block[3]);
 178         pixels += line_size;
 179         block += 8;
 180     }
 181 }
 182
 183 static void add_pixels_clamped2_c(const int16_t *block, uint8_t *restrict pixels,
 184                           int line_size)
 185 {
 186     int i;
 187
 188     /* read the pixels */
 189     for(i=0;i<2;i++) {
 190         pixels[0] = av_clip_uint8(pixels[0] + block[0]);
 191         pixels[1] = av_clip_uint8(pixels[1] + block[1]);
 192         pixels += line_size;
 193         block += 8;
 194     }
 195 }
 196
 197 static void ff_jref_idct4_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 198 {
 199     ff_j_rev_dct4 (block);
 200     put_pixels_clamped4_c(block, dest, line_size);
 201 }
 202 static void ff_jref_idct4_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 203 {
 204     ff_j_rev_dct4 (block);
 205     add_pixels_clamped4_c(block, dest, line_size);
 206 }
 207
 208 static void ff_jref_idct2_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 209 {
 210     ff_j_rev_dct2 (block);
 211     put_pixels_clamped2_c(block, dest, line_size);
 212 }
 213 static void ff_jref_idct2_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 214 {
 215     ff_j_rev_dct2 (block);
 216     add_pixels_clamped2_c(block, dest, line_size);
 217 }
 218
 219 static void ff_jref_idct1_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 220 {
 221     dest[0] = av_clip_uint8((block[0] + 4)>>3);
 222 }
 223 static void ff_jref_idct1_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
 224 {
 225     dest[0] = av_clip_uint8(dest[0] + ((block[0] + 4)>>3));
 226 }
 227
 228 av_cold void ff_idctdsp_init(IDCTDSPContext *c, AVCodecContext *avctx)
 229 {
 230     av_unused const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;
 231
 232     if (avctx->lowres==1) {
 233         c->idct_put  = ff_jref_idct4_put;
 234         c->idct_add  = ff_jref_idct4_add;
 235         c->idct      = ff_j_rev_dct4;
 236         c->perm_type = FF_IDCT_PERM_NONE;
 237     } else if (avctx->lowres==2) {
 238         c->idct_put  = ff_jref_idct2_put;
 239         c->idct_add  = ff_jref_idct2_add;
 240         c->idct      = ff_j_rev_dct2;
 241         c->perm_type = FF_IDCT_PERM_NONE;
 242     } else if (avctx->lowres==3) {
 243         c->idct_put  = ff_jref_idct1_put;
 244         c->idct_add  = ff_jref_idct1_add;
 245         c->idct      = ff_j_rev_dct1;
 246         c->perm_type = FF_IDCT_PERM_NONE;
 247     } else {
 248         if (avctx->bits_per_raw_sample == 10 || avctx->bits_per_raw_sample == 9) {
 249             /* 10-bit MPEG-4 Simple Studio Profile requires a higher precision IDCT
 250                However, it only uses idct_put */
 251             if (c->mpeg4_studio_profile) {
 252                 c->idct_put              = ff_simple_idct_put_int32_10bit;
 253                 c->idct_add              = NULL;
 254                 c->idct                  = NULL;
 255             } else {
 256                 c->idct_put              = ff_simple_idct_put_int16_10bit;
 257                 c->idct_add              = ff_simple_idct_add_int16_10bit;
 258                 c->idct                  = ff_simple_idct_int16_10bit;
 259             }
 260             c->perm_type             = FF_IDCT_PERM_NONE;
 261         } else if (avctx->bits_per_raw_sample == 12) {
 262             c->idct_put              = ff_simple_idct_put_int16_12bit;
 263             c->idct_add              = ff_simple_idct_add_int16_12bit;
 264             c->idct                  = ff_simple_idct_int16_12bit;
 265             c->perm_type             = FF_IDCT_PERM_NONE;
 266         } else {
 267             if (avctx->idct_algo == FF_IDCT_INT) {
 268                 c->idct_put  = ff_jref_idct_put;
 269                 c->idct_add  = ff_jref_idct_add;
 270                 c->idct      = ff_j_rev_dct;
 271                 c->perm_type = FF_IDCT_PERM_LIBMPEG2;
 272 #if CONFIG_FAANIDCT
 273             } else if (avctx->idct_algo == FF_IDCT_FAAN) {
 274                 c->idct_put  = ff_faanidct_put;
 275                 c->idct_add  = ff_faanidct_add;
 276                 c->idct      = ff_faanidct;
 277                 c->perm_type = FF_IDCT_PERM_NONE;
 278 #endif /* CONFIG_FAANIDCT */
 279             } else { // accurate/default
 280                 c->idct_put  = ff_simple_idct_put_int16_8bit;
 281                 c->idct_add  = ff_simple_idct_add_int16_8bit;
 282                 c->idct      = ff_simple_idct_int16_8bit;
 283                 c->perm_type = FF_IDCT_PERM_NONE;
 284             }
 285         }
 286     }
 287
 288     c->put_pixels_clamped        = ff_put_pixels_clamped_c;
 289     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;
 290     c->add_pixels_clamped        = ff_add_pixels_clamped_c;
 291
 292     if (CONFIG_MPEG4_DECODER && avctx->idct_algo == FF_IDCT_XVID)
 293         ff_xvid_idct_init(c, avctx);
 294
 295 #if ARCH_AARCH64
 296     ff_idctdsp_init_aarch64(c, avctx, high_bit_depth);
 297 #elif ARCH_ARM
 298     ff_idctdsp_init_arm(c, avctx, high_bit_depth);
 299 #elif ARCH_PPC
 300     ff_idctdsp_init_ppc(c, avctx, high_bit_depth);
 301 #elif ARCH_RISCV
 302     ff_idctdsp_init_riscv(c, avctx, high_bit_depth);
 303 #elif ARCH_X86
 304     ff_idctdsp_init_x86(c, avctx, high_bit_depth);
 305 #elif ARCH_MIPS
 306     ff_idctdsp_init_mips(c, avctx, high_bit_depth);
 307 #elif ARCH_LOONGARCH
 308     ff_idctdsp_init_loongarch(c, avctx, high_bit_depth);
 309 #endif
 310
 311     ff_init_scantable_permutation(c->idct_permutation,
 312                                   c->perm_type);
 313 }