2 * RV30/40 decoder common dsp functions
3 * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
4 * Copyright (c) 2011 Janne Grunau
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * RV30/40 decoder common dsp functions
31 * @name RV30/40 inverse transform functions
35 static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
39 for(i = 0; i < 4; i++){
40 const int z0 = 13*(block[i+4*0] + block[i+4*2]);
41 const int z1 = 13*(block[i+4*0] - block[i+4*2]);
42 const int z2 = 7* block[i+4*1] - 17*block[i+4*3];
43 const int z3 = 17* block[i+4*1] + 7*block[i+4*3];
45 temp[4*i+0] = z0 + z3;
46 temp[4*i+1] = z1 + z2;
47 temp[4*i+2] = z1 - z2;
48 temp[4*i+3] = z0 - z3;
53 * Real Video 3.0/4.0 inverse transform + sample reconstruction
54 * Code is almost the same as in SVQ3, only scaling is different.
56 static void rv34_idct_add_c(uint8_t *dst, int stride, DCTELEM *block){
58 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
61 rv34_row_transform(temp, block);
62 memset(block, 0, 16*sizeof(DCTELEM));
64 for(i = 0; i < 4; i++){
65 const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
66 const int z1 = 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
67 const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
68 const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
70 dst[0] = cm[ dst[0] + ( (z0 + z3) >> 10 ) ];
71 dst[1] = cm[ dst[1] + ( (z1 + z2) >> 10 ) ];
72 dst[2] = cm[ dst[2] + ( (z1 - z2) >> 10 ) ];
73 dst[3] = cm[ dst[3] + ( (z0 - z3) >> 10 ) ];
80 * RealVideo 3.0/4.0 inverse transform for DC block
82 * Code is almost the same as rv34_inv_transform()
83 * but final coefficients are multiplied by 1.5 and have no rounding.
85 static void rv34_inv_transform_noround_c(DCTELEM *block){
89 rv34_row_transform(temp, block);
91 for(i = 0; i < 4; i++){
92 const int z0 = 13*(temp[4*0+i] + temp[4*2+i]);
93 const int z1 = 13*(temp[4*0+i] - temp[4*2+i]);
94 const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
95 const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
97 block[i*4+0] = ((z0 + z3) * 3) >> 11;
98 block[i*4+1] = ((z1 + z2) * 3) >> 11;
99 block[i*4+2] = ((z1 - z2) * 3) >> 11;
100 block[i*4+3] = ((z0 - z3) * 3) >> 11;
104 static void rv34_idct_dc_add_c(uint8_t *dst, int stride, int dc)
106 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
109 cm += (13*13*dc + 0x200) >> 10;
111 for (i = 0; i < 4; i++)
113 for (j = 0; j < 4; j++)
114 dst[j] = cm[ dst[j] ];
120 static void rv34_inv_transform_dc_noround_c(DCTELEM *block)
122 DCTELEM dc = (13 * 13 * 3 * block[0]) >> 11;
125 for (i = 0; i < 4; i++, block += 4)
126 for (j = 0; j < 4; j++)
130 /** @} */ // transform
133 av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
134 c->rv34_inv_transform = rv34_inv_transform_noround_c;
135 c->rv34_inv_transform_dc = rv34_inv_transform_dc_noround_c;
137 c->rv34_idct_add = rv34_idct_add_c;
138 c->rv34_idct_dc_add = rv34_idct_dc_add_c;
141 ff_rv34dsp_init_neon(c, dsp);
143 ff_rv34dsp_init_x86(c, dsp);