1 /*****************************************************************************
2 * vdec_idct.h : types for the inverse discrete cosine transform
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
5 * $Id: vdec_idct.h,v 1.1 2000/12/21 17:19:52 massiot 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 /*****************************************************************************
31 *****************************************************************************/
33 /*****************************************************************************
35 *****************************************************************************/
38 typedef short dctelem_t;
40 typedef int dctelem_t;
45 #define SPARSE_SCALE_FACTOR 8
46 #define DCTSIZE 8 /* 8*8 DCT */
48 /*****************************************************************************
50 *****************************************************************************/
52 /* We assume that right shift corresponds to signed division by 2 with
53 * rounding towards minus infinity. This is correct for typical "arithmetic
54 * shift" instructions that shift in copies of the sign bit. But some
55 * C compilers implement >> with an unsigned shift. For these machines you
56 * must define RIGHT_SHIFT_IS_UNSIGNED.
57 * RIGHT_SHIFT provides a proper signed right shift of an s32 quantity.
58 * It is only applied with constant shift counts. SHIFT_TEMPS must be
59 * included in the variables of any routine using RIGHT_SHIFT.
62 #ifdef RIGHT_SHIFT_IS_UNSIGNED
63 #define SHIFT_TEMPS s32 shift_temp;
64 #define RIGHT_SHIFT(x,shft) \
65 ((shift_temp = (x)) < 0 ? \
66 (shift_temp >> (shft)) | ((~((s32) 0)) << (32-(shft))) : \
67 (shift_temp >> (shft)))
70 #define RIGHT_SHIFT(x,shft) ((x) >> (shft))
74 * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
75 * on each column. Direct algorithms are also available, but they are
76 * much more complex and seem not to be any faster when reduced to code.
78 * The poop on this scaling stuff is as follows:
80 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
81 * larger than the true IDCT outputs. The final outputs are therefore
82 * a factor of N larger than desired; since N=8 this can be cured by
83 * a simple right shift at the end of the algorithm. The advantage of
84 * this arrangement is that we save two multiplications per 1-D IDCT,
85 * because the y0 and y4 inputs need not be divided by sqrt(N).
87 * We have to do addition and subtraction of the integer inputs, which
88 * is no problem, and multiplication by fractional constants, which is
89 * a problem to do in integer arithmetic. We multiply all the constants
90 * by CONST_SCALE and convert them to integer constants (thus retaining
91 * CONST_BITS bits of precision in the constants). After doing a
92 * multiplication we have to divide the product by CONST_SCALE, with proper
93 * rounding, to produce the correct output. This division can be done
94 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
95 * as long as possible so that partial sums can be added together with
96 * full fractional precision.
98 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
99 * they are represented to better-than-integral precision. These outputs
100 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
101 * with the recommended scaling. (To scale up 12-bit sample data further, an
102 * intermediate s32 array would be needed.)
104 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
105 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
106 * shows that the values given below are the most effective.
109 #define CONST_BITS 8 /* Jimmy chose this constant :) */
111 #ifdef EIGHT_BIT_SAMPLES
114 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
117 #define ONE ((s32) 1)
119 #define CONST_SCALE (ONE << CONST_BITS)
121 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
122 * IMPORTANT: if your compiler doesn't do this arithmetic at compile time,
123 * you will pay a significant penalty in run time. In that case, figure
124 * the correct integer constant values and insert them by hand.
127 #define FIX(x) ((s32) ((x) * CONST_SCALE + 0.5))
129 /* When adding two opposite-signed fixes, the 0.5 cancels */
130 #define FIX2(x) ((s32) ((x) * CONST_SCALE))
132 /* Descale and correctly round an s32 value that's scaled by N bits.
133 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
134 * the fudge factor is correct for either sign of X.
137 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
139 /* Multiply an s32 variable by an s32 constant to yield an s32 result.
140 * For 8-bit samples with the recommended scaling, all the variable
141 * and constant values involved are no more than 16 bits wide, so a
142 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
143 * this provides a useful speedup on many machines.
144 * There is no way to specify a 16x16->32 multiply in portable C, but
145 * some C compilers will do the right thing if you provide the correct
146 * combination of casts.
147 * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
150 #ifdef EIGHT_BIT_SAMPLES
151 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
152 #define MULTIPLY(var,const) (((INT16) (var)) * ((INT16) (const)))
154 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
155 #define MULTIPLY(var,const) (((INT16) (var)) * ((s32) (const)))
159 #ifndef MULTIPLY /* default definition */
160 #define MULTIPLY(var,const) ((var) * (const))
163 /*****************************************************************************
165 *****************************************************************************/
166 typedef void (*f_idct_t)( struct vdec_thread_s *, dctelem_t*, int );
168 /*****************************************************************************
170 *****************************************************************************/
171 void vdec_InitIDCT (struct vdec_thread_s * p_vdec);
172 void vdec_SparseIDCT( struct vdec_thread_s *, dctelem_t*, int );
173 void vdec_IDCT( struct vdec_thread_s *, dctelem_t*, int );