ffmpeg/libavcodec/vp9dsp.h

   1 /*
   2  * VP9 compatible video decoder
   3  *
   4  * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
   5  * Copyright (C) 2013 Clément Bœsch <u pkh me>
   6  *
   7  * This file is part of FFmpeg.
   8  *
   9  * FFmpeg is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * FFmpeg is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with FFmpeg; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #ifndef AVCODEC_VP9DSP_H
  25 #define AVCODEC_VP9DSP_H
  26
  27 #include <stddef.h>
  28 #include <stdint.h>
  29
  30 #include "vp9.h"
  31
  32 typedef void (*vp9_mc_func)(uint8_t *dst, ptrdiff_t dst_stride,
  33                             const uint8_t *ref, ptrdiff_t ref_stride,
  34                             int h, int mx, int my);
  35
  36 typedef struct VP9DSPContext {
  37     /*
  38      * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32
  39      * dimension 2: intra prediction modes
  40      *
  41      * dst/left/top is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
  42      * stride is aligned by 16 pixels
  43      * top[-1] is top/left; top[4,7] is top-right for 4x4
  44      */
  45     // FIXME(rbultje) maybe replace left/top pointers with HAVE_TOP/
  46     // HAVE_LEFT/HAVE_TOPRIGHT flags instead, and then handle it in-place?
  47     // also needs to fit in with what h264/vp8/etc do
  48     void (*intra_pred[N_TXFM_SIZES][N_INTRA_PRED_MODES])(uint8_t *dst,
  49                                                          ptrdiff_t stride,
  50                                                          const uint8_t *left,
  51                                                          const uint8_t *top);
  52
  53     /*
  54      * dimension 1: 0=4x4, 1=8x8, 2=16x16, 3=32x32, 4=lossless (3-4=dct only)
  55      * dimension 2: 0=dct/dct, 1=dct/adst, 2=adst/dct, 3=adst/adst
  56      *
  57      * dst is aligned by transform-size (i.e. 4, 8, 16 or 32 pixels)
  58      * stride is aligned by 16 pixels
  59      * block is 16-byte aligned
  60      * eob indicates the position (+1) of the last non-zero coefficient,
  61      * in scan-order. This can be used to write faster versions, e.g. a
  62      * dc-only 4x4/8x8/16x16/32x32, or a 4x4-only (eob<10) 8x8/16x16/32x32,
  63      * etc.
  64      */
  65     // FIXME also write idct_add_block() versions for whole (inter) pred
  66     // blocks, so we can do 2 4x4s at once
  67     void (*itxfm_add[N_TXFM_SIZES + 1][N_TXFM_TYPES])(uint8_t *dst,
  68                                                       ptrdiff_t stride,
  69                                                       int16_t *block, int eob);
  70
  71     /*
  72      * dimension 1: width of filter (0=4, 1=8, 2=16)
  73      * dimension 2: 0=col-edge filter (h), 1=row-edge filter (v)
  74      *
  75      * dst/stride are aligned by 8
  76      */
  77     void (*loop_filter_8[3][2])(uint8_t *dst, ptrdiff_t stride,
  78                                 int mb_lim, int lim, int hev_thr);
  79
  80     /*
  81      * dimension 1: 0=col-edge filter (h), 1=row-edge filter (v)
  82      *
  83      * The width of filter is assumed to be 16; dst/stride are aligned by 16
  84      */
  85     void (*loop_filter_16[2])(uint8_t *dst, ptrdiff_t stride,
  86                               int mb_lim, int lim, int hev_thr);
  87
  88     /*
  89      * dimension 1/2: width of filter (0=4, 1=8) for each filter half
  90      * dimension 3: 0=col-edge filter (h), 1=row-edge filter (v)
  91      *
  92      * dst/stride are aligned by operation size
  93      * this basically calls loop_filter[d1][d3][0](), followed by
  94      * loop_filter[d2][d3][0]() on the next 8 pixels
  95      * mb_lim/lim/hev_thr contain two values in the lowest two bytes of the
  96      * integer.
  97      */
  98     // FIXME perhaps a mix4 that operates on 32px (for AVX2)
  99     void (*loop_filter_mix2[2][2][2])(uint8_t *dst, ptrdiff_t stride,
 100                                       int mb_lim, int lim, int hev_thr);
 101
 102     /*
 103      * dimension 1: hsize (0: 64, 1: 32, 2: 16, 3: 8, 4: 4)
 104      * dimension 2: filter type (0: smooth, 1: regular, 2: sharp, 3: bilin)
 105      * dimension 3: averaging type (0: put, 1: avg)
 106      * dimension 4: x subpel interpolation (0: none, 1: 8tap/bilin)
 107      * dimension 5: y subpel interpolation (0: none, 1: 8tap/bilin)
 108      *
 109      * dst/stride are aligned by hsize
 110      */
 111     vp9_mc_func mc[5][4][2][2][2];
 112 } VP9DSPContext;
 113
 114 void ff_vp9dsp_init(VP9DSPContext *dsp);
 115
 116 void ff_vp9dsp_init_x86(VP9DSPContext *dsp);
 117
 118 #endif /* AVCODEC_VP9DSP_H */