rv34dsp.c 4 KB
Newer Older
1 2 3 4 5
/*
 * RV30/40 decoder common dsp functions
 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
 * Copyright (c) 2011 Janne Grunau
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9 10 11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14 15 16 17 18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20 21 22 23 24 25 26
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * RV30/40 decoder common dsp functions
 */
27

28
#include "rv34dsp.h"
29
#include "libavutil/common.h"
30 31 32 33 34 35

/**
 * @name RV30/40 inverse transform functions
 * @{
 */

Diego Biurrun's avatar
Diego Biurrun committed
36
static av_always_inline void rv34_row_transform(int temp[16], int16_t *block)
37 38 39 40
{
    int i;

    for(i = 0; i < 4; i++){
41 42 43 44
        const int z0 = 13*(block[i+4*0] +    block[i+4*2]);
        const int z1 = 13*(block[i+4*0] -    block[i+4*2]);
        const int z2 =  7* block[i+4*1] - 17*block[i+4*3];
        const int z3 = 17* block[i+4*1] +  7*block[i+4*3];
45 46 47 48 49 50 51 52

        temp[4*i+0] = z0 + z3;
        temp[4*i+1] = z1 + z2;
        temp[4*i+2] = z1 - z2;
        temp[4*i+3] = z0 - z3;
    }
}

53 54 55 56
/**
 * Real Video 3.0/4.0 inverse transform + sample reconstruction
 * Code is almost the same as in SVQ3, only scaling is different.
 */
Diego Biurrun's avatar
Diego Biurrun committed
57
static void rv34_idct_add_c(uint8_t *dst, ptrdiff_t stride, int16_t *block){
58 59 60 61
    int      temp[16];
    int      i;

    rv34_row_transform(temp, block);
Diego Biurrun's avatar
Diego Biurrun committed
62
    memset(block, 0, 16*sizeof(int16_t));
63 64 65 66 67 68 69

    for(i = 0; i < 4; i++){
        const int z0 = 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;
        const int z1 = 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;
        const int z2 =  7* temp[4*1+i] - 17*temp[4*3+i];
        const int z3 = 17* temp[4*1+i] +  7*temp[4*3+i];

70 71 72 73
        dst[0] = av_clip_uint8( dst[0] + ( (z0 + z3) >> 10 ) );
        dst[1] = av_clip_uint8( dst[1] + ( (z1 + z2) >> 10 ) );
        dst[2] = av_clip_uint8( dst[2] + ( (z1 - z2) >> 10 ) );
        dst[3] = av_clip_uint8( dst[3] + ( (z0 - z3) >> 10 ) );
74 75 76 77 78

        dst  += stride;
    }
}

79 80 81 82 83 84
/**
 * RealVideo 3.0/4.0 inverse transform for DC block
 *
 * Code is almost the same as rv34_inv_transform()
 * but final coefficients are multiplied by 1.5 and have no rounding.
 */
Diego Biurrun's avatar
Diego Biurrun committed
85
static void rv34_inv_transform_noround_c(int16_t *block){
86 87 88 89 90 91
    int temp[16];
    int i;

    rv34_row_transform(temp, block);

    for(i = 0; i < 4; i++){
92 93 94 95 96 97 98 99 100
        const int z0 = 39*(temp[4*0+i] +    temp[4*2+i]);
        const int z1 = 39*(temp[4*0+i] -    temp[4*2+i]);
        const int z2 = 21* temp[4*1+i] - 51*temp[4*3+i];
        const int z3 = 51* temp[4*1+i] + 21*temp[4*3+i];

        block[i*4+0] = (z0 + z3) >> 11;
        block[i*4+1] = (z1 + z2) >> 11;
        block[i*4+2] = (z1 - z2) >> 11;
        block[i*4+3] = (z0 - z3) >> 11;
101 102 103
    }
}

104
static void rv34_idct_dc_add_c(uint8_t *dst, ptrdiff_t stride, int dc)
105 106 107
{
    int i, j;

108
    dc = (13*13*dc + 0x200) >> 10;
109 110 111
    for (i = 0; i < 4; i++)
    {
        for (j = 0; j < 4; j++)
112
            dst[j] = av_clip_uint8( dst[j] + dc );
113 114 115 116 117

        dst += stride;
    }
}

Diego Biurrun's avatar
Diego Biurrun committed
118
static void rv34_inv_transform_dc_noround_c(int16_t *block)
119
{
Diego Biurrun's avatar
Diego Biurrun committed
120
    int16_t dc = (13 * 13 * 3 * block[0]) >> 11;
121 122
    int i, j;

123
    for (i = 0; i < 4; i++, block += 4)
124 125 126 127
        for (j = 0; j < 4; j++)
            block[j] = dc;
}

128 129 130
/** @} */ // transform


131 132
av_cold void ff_rv34dsp_init(RV34DSPContext *c)
{
133 134
    c->rv34_inv_transform    = rv34_inv_transform_noround_c;
    c->rv34_inv_transform_dc = rv34_inv_transform_dc_noround_c;
135

136 137 138
    c->rv34_idct_add    = rv34_idct_add_c;
    c->rv34_idct_dc_add = rv34_idct_dc_add_c;

139
    if (ARCH_ARM)
140
        ff_rv34dsp_init_arm(c);
141
    if (ARCH_X86)
142
        ff_rv34dsp_init_x86(c);
143
}