1 |
2 |
kdv |
/*
|
2 |
|
|
* jrevdct.c
|
3 |
|
|
*
|
4 |
|
|
* Copyright (C) 1991, 1992, Thomas G. Lane.
|
5 |
|
|
* This file is part of the Independent JPEG Group's software.
|
6 |
|
|
* For conditions of distribution and use, see the accompanying README file.
|
7 |
|
|
*
|
8 |
|
|
* This file contains the basic inverse-DCT transformation subroutine.
|
9 |
|
|
*
|
10 |
|
|
* This implementation is based on an algorithm described in
|
11 |
|
|
* C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
|
12 |
|
|
* Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
|
13 |
|
|
* Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
|
14 |
|
|
* The primary algorithm described there uses 11 multiplies and 29 adds.
|
15 |
|
|
* We use their alternate method with 12 multiplies and 32 adds.
|
16 |
|
|
* The advantage of this method is that no data path contains more than one
|
17 |
|
|
* multiplication; this allows a very simple and accurate implementation in
|
18 |
|
|
* scaled fixed-point arithmetic, with a minimal number of shifts.
|
19 |
|
|
*/
|
20 |
|
|
|
21 |
|
|
#include "dct.h"
|
22 |
|
|
#include <stdio.h>
|
23 |
|
|
|
24 |
|
|
/* We assume that right shift corresponds to signed division by 2 with
|
25 |
|
|
* rounding towards minus infinity. This is correct for typical "arithmetic
|
26 |
|
|
* shift" instructions that shift in copies of the sign bit. But some
|
27 |
|
|
* C compilers implement >> with an unsigned shift. For these machines you
|
28 |
|
|
* must define RIGHT_SHIFT_IS_UNSIGNED.
|
29 |
|
|
* RIGHT_SHIFT provides a proper signed right shift of an INT32 quantity.
|
30 |
|
|
* It is only applied with constant shift counts. SHIFT_TEMPS must be
|
31 |
|
|
* included in the variables of any routine using RIGHT_SHIFT.
|
32 |
|
|
*/
|
33 |
|
|
|
34 |
|
|
#ifdef RIGHT_SHIFT_IS_UNSIGNED
|
35 |
|
|
#define SHIFT_TEMPS INT32 shift_temp;
|
36 |
|
|
#define RIGHT_SHIFT(x,shft) \
|
37 |
|
|
((shift_temp = (x)) < 0 ? \
|
38 |
|
|
(shift_temp >> (shft)) | ((~((INT32) 0)) << (32-(shft))) : \
|
39 |
|
|
(shift_temp >> (shft)))
|
40 |
|
|
#else
|
41 |
|
|
#define SHIFT_TEMPS
|
42 |
|
|
#define RIGHT_SHIFT(x,shft) ((x) >> (shft))
|
43 |
|
|
#endif
|
44 |
|
|
|
45 |
|
|
|
46 |
|
|
/*
|
47 |
|
|
* This routine is specialized to the case DCTSIZE = 8.
|
48 |
|
|
*/
|
49 |
|
|
|
50 |
|
|
#if DCTSIZE != 8
|
51 |
|
|
Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
|
52 |
|
|
#endif
|
53 |
|
|
|
54 |
|
|
|
55 |
|
|
/*
|
56 |
|
|
* A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
|
57 |
|
|
* on each column. Direct algorithms are also available, but they are
|
58 |
|
|
* much more complex and seem not to be any faster when reduced to code.
|
59 |
|
|
*
|
60 |
|
|
* The poop on this scaling stuff is as follows:
|
61 |
|
|
*
|
62 |
|
|
* Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
|
63 |
|
|
* larger than the true IDCT outputs. The final outputs are therefore
|
64 |
|
|
* a factor of N larger than desired; since N=8 this can be cured by
|
65 |
|
|
* a simple right shift at the end of the algorithm. The advantage of
|
66 |
|
|
* this arrangement is that we save two multiplications per 1-D IDCT,
|
67 |
|
|
* because the y0 and y4 inputs need not be divided by sqrt(N).
|
68 |
|
|
*
|
69 |
|
|
* We have to do addition and subtraction of the integer inputs, which
|
70 |
|
|
* is no problem, and multiplication by fractional constants, which is
|
71 |
|
|
* a problem to do in integer arithmetic. We multiply all the constants
|
72 |
|
|
* by CONST_SCALE and convert them to integer constants (thus retaining
|
73 |
|
|
* CONST_BITS bits of precision in the constants). After doing a
|
74 |
|
|
* multiplication we have to divide the product by CONST_SCALE, with proper
|
75 |
|
|
* rounding, to produce the correct output. This division can be done
|
76 |
|
|
* cheaply as a right shift of CONST_BITS bits. We postpone shifting
|
77 |
|
|
* as long as possible so that partial sums can be added together with
|
78 |
|
|
* full fractional precision.
|
79 |
|
|
*
|
80 |
|
|
* The outputs of the first pass are scaled up by PASS1_BITS bits so that
|
81 |
|
|
* they are represented to better-than-integral precision. These outputs
|
82 |
|
|
* require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
|
83 |
|
|
* with the recommended scaling. (To scale up 12-bit sample data further, an
|
84 |
|
|
* intermediate INT32 array would be needed.)
|
85 |
|
|
*
|
86 |
|
|
* To avoid overflow of the 32-bit intermediate results in pass 2, we must
|
87 |
|
|
* have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
|
88 |
|
|
* shows that the values given below are the most effective.
|
89 |
|
|
*/
|
90 |
|
|
|
91 |
|
|
#ifdef EIGHT_BIT_SAMPLES
|
92 |
|
|
#define CONST_BITS 13
|
93 |
|
|
#define PASS1_BITS 2
|
94 |
|
|
#else
|
95 |
|
|
#define CONST_BITS 13
|
96 |
|
|
#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
|
97 |
|
|
#endif
|
98 |
|
|
|
99 |
|
|
#define ONE ((INT32) 1)
|
100 |
|
|
|
101 |
|
|
#define CONST_SCALE (ONE << CONST_BITS)
|
102 |
|
|
|
103 |
|
|
/* Convert a positive real constant to an integer scaled by CONST_SCALE. */
|
104 |
|
|
|
105 |
|
|
#define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5))
|
106 |
|
|
|
107 |
|
|
/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
|
108 |
|
|
* causing a lot of useless floating-point operations at run time.
|
109 |
|
|
* To get around this we use the following pre-calculated constants.
|
110 |
|
|
* If you change CONST_BITS you may want to add appropriate values.
|
111 |
|
|
* (With a reasonable C compiler, you can just rely on the FIX() macro...)
|
112 |
|
|
*/
|
113 |
|
|
|
114 |
|
|
#if CONST_BITS == 13
|
115 |
|
|
#define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */
|
116 |
|
|
#define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */
|
117 |
|
|
#define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */
|
118 |
|
|
#define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
|
119 |
|
|
#define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
|
120 |
|
|
#define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */
|
121 |
|
|
#define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */
|
122 |
|
|
#define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
|
123 |
|
|
#define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */
|
124 |
|
|
#define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */
|
125 |
|
|
#define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
|
126 |
|
|
#define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */
|
127 |
|
|
#else
|
128 |
|
|
#define FIX_0_298631336 FIX(0.298631336)
|
129 |
|
|
#define FIX_0_390180644 FIX(0.390180644)
|
130 |
|
|
#define FIX_0_541196100 FIX(0.541196100)
|
131 |
|
|
#define FIX_0_765366865 FIX(0.765366865)
|
132 |
|
|
#define FIX_0_899976223 FIX(0.899976223)
|
133 |
|
|
#define FIX_1_175875602 FIX(1.175875602)
|
134 |
|
|
#define FIX_1_501321110 FIX(1.501321110)
|
135 |
|
|
#define FIX_1_847759065 FIX(1.847759065)
|
136 |
|
|
#define FIX_1_961570560 FIX(1.961570560)
|
137 |
|
|
#define FIX_2_053119869 FIX(2.053119869)
|
138 |
|
|
#define FIX_2_562915447 FIX(2.562915447)
|
139 |
|
|
#define FIX_3_072711026 FIX(3.072711026)
|
140 |
|
|
#endif
|
141 |
|
|
|
142 |
|
|
|
143 |
|
|
/* Descale and correctly round an INT32 value that's scaled by N bits.
|
144 |
|
|
* We assume RIGHT_SHIFT rounds towards minus infinity, so adding
|
145 |
|
|
* the fudge factor is correct for either sign of X.
|
146 |
|
|
*/
|
147 |
|
|
|
148 |
|
|
#define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
|
149 |
|
|
|
150 |
|
|
/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
|
151 |
|
|
* For 8-bit samples with the recommended scaling, all the variable
|
152 |
|
|
* and constant values involved are no more than 16 bits wide, so a
|
153 |
|
|
* 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
|
154 |
|
|
* this provides a useful speedup on many machines.
|
155 |
|
|
* There is no way to specify a 16x16->32 multiply in portable C, but
|
156 |
|
|
* some C compilers will do the right thing if you provide the correct
|
157 |
|
|
* combination of casts.
|
158 |
|
|
* NB: for 12-bit samples, a full 32-bit multiplication will be needed.
|
159 |
|
|
*/
|
160 |
|
|
|
161 |
|
|
#ifdef EIGHT_BIT_SAMPLES
|
162 |
|
|
#ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
|
163 |
|
|
#define MULTIPLY(var,const) (((INT16) (var)) * ((INT16) (const)))
|
164 |
|
|
#endif
|
165 |
|
|
#ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
|
166 |
|
|
#define MULTIPLY(var,const) (((INT16) (var)) * ((INT32) (const)))
|
167 |
|
|
#endif
|
168 |
|
|
#endif
|
169 |
|
|
|
170 |
|
|
#ifndef MULTIPLY /* default definition */
|
171 |
|
|
#define MULTIPLY(var,const) ((var) * (const))
|
172 |
|
|
#endif
|
173 |
|
|
|
174 |
|
|
|
175 |
|
|
/*
|
176 |
|
|
* Perform the inverse DCT on one block of coefficients.
|
177 |
|
|
*/
|
178 |
|
|
|
179 |
|
|
void
|
180 |
|
|
j_rev_dct (DCTBLOCK data)
|
181 |
|
|
{
|
182 |
|
|
register DCTELEM *dataptr;
|
183 |
|
|
int i, j;
|
184 |
|
|
FILE *idctdat;
|
185 |
|
|
SHIFT_TEMPS
|
186 |
|
|
|
187 |
|
|
|
188 |
|
|
dataptr = data;
|
189 |
|
|
idctdat = fopen("idct-in", "w");
|
190 |
|
|
for (i = 0; i <= 63; i++) {
|
191 |
|
|
j = ((INT32)dataptr[i]) & 4095;
|
192 |
|
|
fprintf(idctdat, "%x\n", j);
|
193 |
|
|
}
|
194 |
|
|
fclose(idctdat);
|
195 |
|
|
system("./idct-verilog > idct-out");
|
196 |
|
|
idctdat = fopen("idct-out", "r");
|
197 |
|
|
for (i = 0; i <= 63; i++) {
|
198 |
|
|
fscanf(idctdat, "%d", &j);
|
199 |
|
|
dataptr[i] = j;
|
200 |
|
|
}
|
201 |
|
|
fclose(idctdat);
|
202 |
|
|
}
|