1 /*
2  * jdct.h
3  *
4  * Copyright (C) 1994-1996, 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 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.
26  * Quantization of the output coefficients is done by jcdctmgr.c.
27  */
28 
29 #if BITS_IN_JSAMPLE == 8
30 #ifdef ANDROID_MIPS_IDCT
31 typedef short DCTELEM;		/* 16 or 32 bits is fine */
32 #else
33 typedef int DCTELEM;		/* 16 or 32 bits is fine */
34 #endif
35 #else
36 typedef INT32 DCTELEM;		/* must have 32 bits */
37 #endif
38 
39 typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data));
40 typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data));
41 
42 
43 /*
44  * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
45  * to an output sample array.  The routine must dequantize the input data as
46  * well as perform the IDCT; for dequantization, it uses the multiplier table
47  * pointed to by compptr->dct_table.  The output data is to be placed into the
48  * sample array starting at a specified column.  (Any row offset needed will
49  * be applied to the array pointer before it is passed to the IDCT code.)
50  * Note that the number of samples emitted by the IDCT routine is
51  * DCT_scaled_size * DCT_scaled_size.
52  */
53 
54 /* typedef inverse_DCT_method_ptr is declared in jpegint.h */
55 
56 /*
57  * Each IDCT routine has its own ideas about the best dct_table element type.
58  */
59 
60 typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
61 #if BITS_IN_JSAMPLE == 8
62 typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
63 #define IFAST_SCALE_BITS  2	/* fractional bits in scale factors */
64 #else
65 typedef INT32 IFAST_MULT_TYPE;	/* need 32 bits for scaled quantizers */
66 #define IFAST_SCALE_BITS  13	/* fractional bits in scale factors */
67 #endif
68 typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
69 
70 
71 /*
72  * Each IDCT routine is responsible for range-limiting its results and
73  * converting them to unsigned form (0..MAXJSAMPLE).  The raw outputs could
74  * be quite far out of range if the input data is corrupt, so a bulletproof
75  * range-limiting step is required.  We use a mask-and-table-lookup method
76  * to do the combined operations quickly.  See the comments with
77  * prepare_range_limit_table (in jdmaster.c) for more info.
78  */
79 
80 #define IDCT_range_limit(cinfo)  ((cinfo)->sample_range_limit + CENTERJSAMPLE)
81 
82 #define RANGE_MASK  (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
83 
84 
85 /* Short forms of external names for systems with brain-damaged linkers. */
86 
87 #ifdef NEED_SHORT_EXTERNAL_NAMES
88 #define jpeg_fdct_islow		jFDislow
89 #define jpeg_fdct_ifast		jFDifast
90 #define jpeg_fdct_float		jFDfloat
91 #define jpeg_idct_islow		jRDislow
92 #define jpeg_idct_ifast		jRDifast
93 #define jpeg_idct_float		jRDfloat
94 #define jpeg_idct_4x4		jRD4x4
95 #define jpeg_idct_2x2		jRD2x2
96 #define jpeg_idct_1x1		jRD1x1
97 #endif /* NEED_SHORT_EXTERNAL_NAMES */
98 
99 /* Extern declarations for the forward and inverse DCT routines. */
100 
101 EXTERN(void) jpeg_fdct_islow JPP((DCTELEM * data));
102 EXTERN(void) jpeg_fdct_ifast JPP((DCTELEM * data));
103 EXTERN(void) jpeg_fdct_float JPP((FAST_FLOAT * data));
104 
105 EXTERN(void) jpeg_idct_islow
106     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
107 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
108 EXTERN(void) jpeg_idct_ifast
109     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
110 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
111 EXTERN(void) jpeg_idct_float
112     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
113 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
114 EXTERN(void) jpeg_idct_4x4
115     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
116 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
117 EXTERN(void) jpeg_idct_2x2
118     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
119 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
120 EXTERN(void) jpeg_idct_1x1
121     JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
122 	 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
123 
124 
125 /*
126  * Macros for handling fixed-point arithmetic; these are used by many
127  * but not all of the DCT/IDCT modules.
128  *
129  * All values are expected to be of type INT32.
130  * Fractional constants are scaled left by CONST_BITS bits.
131  * CONST_BITS is defined within each module using these macros,
132  * and may differ from one module to the next.
133  */
134 
135 #define ONE	((INT32) 1)
136 #define CONST_SCALE (ONE << CONST_BITS)
137 
138 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
139  * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
140  * thus causing a lot of useless floating-point operations at run time.
141  */
142 
143 #define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))
144 
145 /* Descale and correctly round an INT32 value that's scaled by N bits.
146  * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
147  * the fudge factor is correct for either sign of X.
148  */
149 
150 #define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
151 
152 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
153  * This macro is used only when the two inputs will actually be no more than
154  * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
155  * full 32x32 multiply.  This provides a useful speedup on many machines.
156  * Unfortunately there is no way to specify a 16x16->32 multiply portably
157  * in C, but some C compilers will do the right thing if you provide the
158  * correct combination of casts.
159  */
160 
161 #ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
162 #define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT16) (const)))
163 #endif
164 #ifdef SHORTxLCONST_32		/* known to work with Microsoft C 6.0 */
165 #define MULTIPLY16C16(var,const)  (((INT16) (var)) * ((INT32) (const)))
166 #endif
167 
168 #ifndef MULTIPLY16C16		/* default definition */
169 #define MULTIPLY16C16(var,const)  ((var) * (const))
170 #endif
171 
172 /* Same except both inputs are variables. */
173 
174 #ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
175 #define MULTIPLY16V16(var1,var2)  (((INT16) (var1)) * ((INT16) (var2)))
176 #endif
177 
178 #ifndef MULTIPLY16V16		/* default definition */
179 #define MULTIPLY16V16(var1,var2)  ((var1) * (var2))
180 #endif
181