1 /*****************************************************************************
2 * vdec_idct.h : types for the inverse discrete cosine transform
3 *****************************************************************************
4 * Copyright (C) 1999, 2000 VideoLAN
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program 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 * General Public License for more details.
18 * You should have received a copy of the GNU General Public
19 * License along with this program; if not, write to the
20 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
21 * Boston, MA 02111-1307, USA.
22 *****************************************************************************/
24 /*****************************************************************************
30 *****************************************************************************/
32 /*****************************************************************************
34 *****************************************************************************/
37 typedef short dctelem_t;
39 typedef int dctelem_t;
44 #define SPARSE_SCALE_FACTOR 8
45 #define DCTSIZE 8 /* 8*8 DCT */
47 /*****************************************************************************
49 *****************************************************************************/
51 /* We assume that right shift corresponds to signed division by 2 with
52 * rounding towards minus infinity. This is correct for typical "arithmetic
53 * shift" instructions that shift in copies of the sign bit. But some
54 * C compilers implement >> with an unsigned shift. For these machines you
55 * must define RIGHT_SHIFT_IS_UNSIGNED.
56 * RIGHT_SHIFT provides a proper signed right shift of an s32 quantity.
57 * It is only applied with constant shift counts. SHIFT_TEMPS must be
58 * included in the variables of any routine using RIGHT_SHIFT.
61 #ifdef RIGHT_SHIFT_IS_UNSIGNED
62 #define SHIFT_TEMPS s32 shift_temp;
63 #define RIGHT_SHIFT(x,shft) \
64 ((shift_temp = (x)) < 0 ? \
65 (shift_temp >> (shft)) | ((~((s32) 0)) << (32-(shft))) : \
66 (shift_temp >> (shft)))
69 #define RIGHT_SHIFT(x,shft) ((x) >> (shft))
73 * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
74 * on each column. Direct algorithms are also available, but they are
75 * much more complex and seem not to be any faster when reduced to code.
77 * The poop on this scaling stuff is as follows:
79 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
80 * larger than the true IDCT outputs. The final outputs are therefore
81 * a factor of N larger than desired; since N=8 this can be cured by
82 * a simple right shift at the end of the algorithm. The advantage of
83 * this arrangement is that we save two multiplications per 1-D IDCT,
84 * because the y0 and y4 inputs need not be divided by sqrt(N).
86 * We have to do addition and subtraction of the integer inputs, which
87 * is no problem, and multiplication by fractional constants, which is
88 * a problem to do in integer arithmetic. We multiply all the constants
89 * by CONST_SCALE and convert them to integer constants (thus retaining
90 * CONST_BITS bits of precision in the constants). After doing a
91 * multiplication we have to divide the product by CONST_SCALE, with proper
92 * rounding, to produce the correct output. This division can be done
93 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
94 * as long as possible so that partial sums can be added together with
95 * full fractional precision.
97 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
98 * they are represented to better-than-integral precision. These outputs
99 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
100 * with the recommended scaling. (To scale up 12-bit sample data further, an
101 * intermediate s32 array would be needed.)
103 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
104 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
105 * shows that the values given below are the most effective.
108 #define CONST_BITS 8 /* Jimmy chose this constant :) */
110 #ifdef EIGHT_BIT_SAMPLES
113 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
116 #define ONE ((s32) 1)
118 #define CONST_SCALE (ONE << CONST_BITS)
120 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
121 * IMPORTANT: if your compiler doesn't do this arithmetic at compile time,
122 * you will pay a significant penalty in run time. In that case, figure
123 * the correct integer constant values and insert them by hand.
126 #define FIX(x) ((s32) ((x) * CONST_SCALE + 0.5))
128 /* When adding two opposite-signed fixes, the 0.5 cancels */
129 #define FIX2(x) ((s32) ((x) * CONST_SCALE))
131 /* Descale and correctly round an s32 value that's scaled by N bits.
132 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
133 * the fudge factor is correct for either sign of X.
136 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
138 /* Multiply an s32 variable by an s32 constant to yield an s32 result.
139 * For 8-bit samples with the recommended scaling, all the variable
140 * and constant values involved are no more than 16 bits wide, so a
141 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
142 * this provides a useful speedup on many machines.
143 * There is no way to specify a 16x16->32 multiply in portable C, but
144 * some C compilers will do the right thing if you provide the correct
145 * combination of casts.
146 * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
149 #ifdef EIGHT_BIT_SAMPLES
150 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
151 #define MULTIPLY(var,const) (((INT16) (var)) * ((INT16) (const)))
153 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
154 #define MULTIPLY(var,const) (((INT16) (var)) * ((s32) (const)))
158 #ifndef MULTIPLY /* default definition */
159 #define MULTIPLY(var,const) ((var) * (const))
162 /*****************************************************************************
164 *****************************************************************************/
165 typedef void (*f_idct_t)( struct vdec_thread_s *, dctelem_t*, int );
167 /*****************************************************************************
169 *****************************************************************************/
170 void vdec_InitIDCT (struct vdec_thread_s * p_vdec);
171 void vdec_SparseIDCT( struct vdec_thread_s *, dctelem_t*, int );
172 void vdec_IDCT( struct vdec_thread_s *, dctelem_t*, int );