blob: 7b49a97526d0a285e71eaeb00a7b1601d137bdf5 [file] [log] [blame]
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +00001/*
2 * jdct.h
3 *
Thomas G. Lane489583f1996-02-07 00:00:00 +00004 * Copyright (C) 1994-1996, Thomas G. Lane.
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +00005 * 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 include file contains common declarations for the forward and
9 * inverse DCT modules. These declarations are private to the DCT managers
10 * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
11 * The individual DCT algorithms are kept in separate files to ease
12 * machine-dependent tuning (e.g., assembly coding).
13 */
14
15
16/*
17 * A forward DCT routine is given a pointer to a work area of type DCTELEM[];
18 * the DCT is to be performed in-place in that buffer. Type DCTELEM is int
19 * for 8-bit samples, INT32 for 12-bit samples. (NOTE: Floating-point DCT
20 * implementations use an array of type FAST_FLOAT, instead.)
21 * The DCT inputs are expected to be signed (range +-CENTERJSAMPLE).
22 * The DCT outputs are returned scaled up by a factor of 8; they therefore
23 * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This
24 * convention improves accuracy in integer implementations and saves some
25 * work in floating-point ones.
Pierre Ossmandedc42e2009-03-09 13:23:04 +000026 * Quantization of the output coefficients is done by jcdctmgr.c. This
27 * step requires an unsigned type and also one with twice the bits.
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000028 */
29
30#if BITS_IN_JSAMPLE == 8
Pierre Ossman5eb84ff2009-03-09 13:25:30 +000031#ifndef WITH_SIMD
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000032typedef int DCTELEM; /* 16 or 32 bits is fine */
Pierre Ossmandedc42e2009-03-09 13:23:04 +000033typedef unsigned int UDCTELEM;
34typedef unsigned long long UDCTELEM2;
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000035#else
Pierre Ossman5eb84ff2009-03-09 13:25:30 +000036typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */
37typedef unsigned short UDCTELEM;
38typedef unsigned int UDCTELEM2;
39#endif
40#else
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000041typedef INT32 DCTELEM; /* must have 32 bits */
Pierre Ossmandedc42e2009-03-09 13:23:04 +000042typedef UINT32 UDCTELEM;
43typedef unsigned long long UDCTELEM2;
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000044#endif
45
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +000046
47/*
48 * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
49 * to an output sample array. The routine must dequantize the input data as
50 * well as perform the IDCT; for dequantization, it uses the multiplier table
51 * pointed to by compptr->dct_table. The output data is to be placed into the
52 * sample array starting at a specified column. (Any row offset needed will
53 * be applied to the array pointer before it is passed to the IDCT code.)
54 * Note that the number of samples emitted by the IDCT routine is
55 * DCT_scaled_size * DCT_scaled_size.
56 */
57
58/* typedef inverse_DCT_method_ptr is declared in jpegint.h */
59
60/*
61 * Each IDCT routine has its own ideas about the best dct_table element type.
62 */
63
64typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
65#if BITS_IN_JSAMPLE == 8
66typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
67#define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */
68#else
69typedef INT32 IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */
70#define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */
71#endif
72typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
73
74
75/*
76 * Each IDCT routine is responsible for range-limiting its results and
77 * converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could
78 * be quite far out of range if the input data is corrupt, so a bulletproof
79 * range-limiting step is required. We use a mask-and-table-lookup method
80 * to do the combined operations quickly. See the comments with
81 * prepare_range_limit_table (in jdmaster.c) for more info.
82 */
83
84#define IDCT_range_limit(cinfo) ((cinfo)->sample_range_limit + CENTERJSAMPLE)
85
86#define RANGE_MASK (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
87
88
89/* Short forms of external names for systems with brain-damaged linkers. */
90
91#ifdef NEED_SHORT_EXTERNAL_NAMES
92#define jpeg_fdct_islow jFDislow
93#define jpeg_fdct_ifast jFDifast
94#define jpeg_fdct_float jFDfloat
95#define jpeg_idct_islow jRDislow
96#define jpeg_idct_ifast jRDifast
97#define jpeg_idct_float jRDfloat
98#define jpeg_idct_4x4 jRD4x4
99#define jpeg_idct_2x2 jRD2x2
100#define jpeg_idct_1x1 jRD1x1
101#endif /* NEED_SHORT_EXTERNAL_NAMES */
102
103/* Extern declarations for the forward and inverse DCT routines. */
104
Thomas G. Lane489583f1996-02-07 00:00:00 +0000105EXTERN(void) jpeg_fdct_islow JPP((DCTELEM * data));
106EXTERN(void) jpeg_fdct_ifast JPP((DCTELEM * data));
107EXTERN(void) jpeg_fdct_float JPP((FAST_FLOAT * data));
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000108
Thomas G. Lane489583f1996-02-07 00:00:00 +0000109EXTERN(void) jpeg_idct_islow
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000110 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
111 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
Thomas G. Lane489583f1996-02-07 00:00:00 +0000112EXTERN(void) jpeg_idct_ifast
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000113 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
114 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
Thomas G. Lane489583f1996-02-07 00:00:00 +0000115EXTERN(void) jpeg_idct_float
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000116 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
117 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
Thomas G. Lane489583f1996-02-07 00:00:00 +0000118EXTERN(void) jpeg_idct_4x4
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000119 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
120 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
Thomas G. Lane489583f1996-02-07 00:00:00 +0000121EXTERN(void) jpeg_idct_2x2
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000122 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
123 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
Thomas G. Lane489583f1996-02-07 00:00:00 +0000124EXTERN(void) jpeg_idct_1x1
Thomas G. Lane36a4ccc1994-09-24 00:00:00 +0000125 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
126 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
127
128
129/*
130 * Macros for handling fixed-point arithmetic; these are used by many
131 * but not all of the DCT/IDCT modules.
132 *
133 * All values are expected to be of type INT32.
134 * Fractional constants are scaled left by CONST_BITS bits.
135 * CONST_BITS is defined within each module using these macros,
136 * and may differ from one module to the next.
137 */
138
139#define ONE ((INT32) 1)
140#define CONST_SCALE (ONE << CONST_BITS)
141
142/* Convert a positive real constant to an integer scaled by CONST_SCALE.
143 * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
144 * thus causing a lot of useless floating-point operations at run time.
145 */
146
147#define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5))
148
149/* Descale and correctly round an INT32 value that's scaled by N bits.
150 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
151 * the fudge factor is correct for either sign of X.
152 */
153
154#define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
155
156/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
157 * This macro is used only when the two inputs will actually be no more than
158 * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
159 * full 32x32 multiply. This provides a useful speedup on many machines.
160 * Unfortunately there is no way to specify a 16x16->32 multiply portably
161 * in C, but some C compilers will do the right thing if you provide the
162 * correct combination of casts.
163 */
164
165#ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
166#define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT16) (const)))
167#endif
168#ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
169#define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT32) (const)))
170#endif
171
172#ifndef MULTIPLY16C16 /* default definition */
173#define MULTIPLY16C16(var,const) ((var) * (const))
174#endif
175
176/* Same except both inputs are variables. */
177
178#ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
179#define MULTIPLY16V16(var1,var2) (((INT16) (var1)) * ((INT16) (var2)))
180#endif
181
182#ifndef MULTIPLY16V16 /* default definition */
183#define MULTIPLY16V16(var1,var2) ((var1) * (var2))
184#endif