1 /*****************************************************************************
2 * idct.h : macros for the inverse discrete cosine transform
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: idct.h,v 1.3 2001/04/15 04:19:57 sam Exp $
7 * Authors: Gaƫl Hendryckx <jimmy@via.ecp.fr>
8 * Christophe Massiot <massiot@via.ecp.fr>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
23 *****************************************************************************/
25 #define SPARSE_SCALE_FACTOR 8
26 #define DCTSIZE 8 /* 8*8 DCT */
28 /*****************************************************************************
30 *****************************************************************************/
32 /* We assume that right shift corresponds to signed division by 2 with
33 * rounding towards minus infinity. This is correct for typical "arithmetic
34 * shift" instructions that shift in copies of the sign bit. But some
35 * C compilers implement >> with an unsigned shift. For these machines you
36 * must define RIGHT_SHIFT_IS_UNSIGNED.
37 * RIGHT_SHIFT provides a proper signed right shift of an s32 quantity.
38 * It is only applied with constant shift counts. SHIFT_TEMPS must be
39 * included in the variables of any routine using RIGHT_SHIFT.
42 #ifdef RIGHT_SHIFT_IS_UNSIGNED
43 #define SHIFT_TEMPS s32 shift_temp;
44 #define RIGHT_SHIFT(x,shft) \
45 ((shift_temp = (x)) < 0 ? \
46 (shift_temp >> (shft)) | ((~((s32) 0)) << (32-(shft))) : \
47 (shift_temp >> (shft)))
50 #define RIGHT_SHIFT(x,shft) ((x) >> (shft))
54 * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
55 * on each column. Direct algorithms are also available, but they are
56 * much more complex and seem not to be any faster when reduced to code.
58 * The poop on this scaling stuff is as follows:
60 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
61 * larger than the true IDCT outputs. The final outputs are therefore
62 * a factor of N larger than desired; since N=8 this can be cured by
63 * a simple right shift at the end of the algorithm. The advantage of
64 * this arrangement is that we save two multiplications per 1-D IDCT,
65 * because the y0 and y4 inputs need not be divided by sqrt(N).
67 * We have to do addition and subtraction of the integer inputs, which
68 * is no problem, and multiplication by fractional constants, which is
69 * a problem to do in integer arithmetic. We multiply all the constants
70 * by CONST_SCALE and convert them to integer constants (thus retaining
71 * CONST_BITS bits of precision in the constants). After doing a
72 * multiplication we have to divide the product by CONST_SCALE, with proper
73 * rounding, to produce the correct output. This division can be done
74 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
75 * as long as possible so that partial sums can be added together with
76 * full fractional precision.
78 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
79 * they are represented to better-than-integral precision. These outputs
80 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
81 * with the recommended scaling. (To scale up 12-bit sample data further, an
82 * intermediate s32 array would be needed.)
84 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
85 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
86 * shows that the values given below are the most effective.
89 #define CONST_BITS 8 /* Jimmy chose this constant :) */
91 #ifdef EIGHT_BIT_SAMPLES
94 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
99 #define CONST_SCALE (ONE << CONST_BITS)
101 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
102 * IMPORTANT: if your compiler doesn't do this arithmetic at compile time,
103 * you will pay a significant penalty in run time. In that case, figure
104 * the correct integer constant values and insert them by hand.
107 #define FIX(x) ((s32) ((x) * CONST_SCALE + 0.5))
109 /* When adding two opposite-signed fixes, the 0.5 cancels */
110 #define FIX2(x) ((s32) ((x) * CONST_SCALE))
112 /* Descale and correctly round an s32 value that's scaled by N bits.
113 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
114 * the fudge factor is correct for either sign of X.
117 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
119 /* Multiply an s32 variable by an s32 constant to yield an s32 result.
120 * For 8-bit samples with the recommended scaling, all the variable
121 * and constant values involved are no more than 16 bits wide, so a
122 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
123 * this provides a useful speedup on many machines.
124 * There is no way to specify a 16x16->32 multiply in portable C, but
125 * some C compilers will do the right thing if you provide the correct
126 * combination of casts.
127 * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
130 #ifdef EIGHT_BIT_SAMPLES
131 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
132 #define MULTIPLY(var,const) (((INT16) (var)) * ((INT16) (const)))
134 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
135 #define MULTIPLY(var,const) (((INT16) (var)) * ((s32) (const)))
139 #ifndef MULTIPLY /* default definition */
140 #define MULTIPLY(var,const) ((var) * (const))
143 /*****************************************************************************
145 *****************************************************************************/
146 void _M( vdec_SparseIDCT ) ( vdec_thread_t * p_vdec, dctelem_t * p_block,
148 void _M( vdec_InitIDCT ) ( vdec_thread_t * p_vdec );
149 void _M( vdec_IDCT ) ( vdec_thread_t * p_vdec, dctelem_t * p_block,