4 * This file was part of the Independent JPEG Group's software:
5 * Copyright (C) 1994-1996, Thomas G. Lane.
6 * libjpeg-turbo Modifications:
7 * Copyright (C) 2015, 2022, D. R. Commander.
8 * For conditions of distribution and use, see the accompanying README.ijg
11 * This include file contains common declarations for the forward and
12 * inverse DCT modules. These declarations are private to the DCT managers
13 * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
14 * The individual DCT algorithms are kept in separate files to ease
15 * machine-dependent tuning (e.g., assembly coding).
18 #include "jsamplecomp.h"
22 * A forward DCT routine is given a pointer to a work area of type DCTELEM[];
23 * the DCT is to be performed in-place in that buffer. Type DCTELEM is int
24 * for 8-bit samples, JLONG for 12-bit samples. (NOTE: Floating-point DCT
25 * implementations use an array of type FAST_FLOAT, instead.)
26 * The DCT inputs are expected to be signed (range +-_CENTERJSAMPLE).
27 * The DCT outputs are returned scaled up by a factor of 8; they therefore
28 * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This
29 * convention improves accuracy in integer implementations and saves some
30 * work in floating-point ones.
31 * Quantization of the output coefficients is done by jcdctmgr.c. This
32 * step requires an unsigned type and also one with twice the bits.
35 #if BITS_IN_JSAMPLE == 8
37 typedef int DCTELEM; /* 16 or 32 bits is fine */
38 typedef unsigned int UDCTELEM;
39 typedef unsigned long long UDCTELEM2;
41 typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */
42 typedef unsigned short UDCTELEM;
43 typedef unsigned int UDCTELEM2;
46 typedef JLONG DCTELEM; /* must have 32 bits */
47 typedef unsigned long long UDCTELEM2;
52 * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
53 * to an output sample array. The routine must dequantize the input data as
54 * well as perform the IDCT; for dequantization, it uses the multiplier table
55 * pointed to by compptr->dct_table. The output data is to be placed into the
56 * sample array starting at a specified column. (Any row offset needed will
57 * be applied to the array pointer before it is passed to the IDCT code.)
58 * Note that the number of samples emitted by the IDCT routine is
59 * DCT_scaled_size * DCT_scaled_size.
62 /* typedef inverse_DCT_method_ptr is declared in jpegint.h */
65 * Each IDCT routine has its own ideas about the best dct_table element type.
68 typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
69 #if BITS_IN_JSAMPLE == 8
70 typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
71 #define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */
73 typedef JLONG IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */
74 #define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */
76 typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
80 * Each IDCT routine is responsible for range-limiting its results and
81 * converting them to unsigned form (0.._MAXJSAMPLE). The raw outputs could
82 * be quite far out of range if the input data is corrupt, so a bulletproof
83 * range-limiting step is required. We use a mask-and-table-lookup method
84 * to do the combined operations quickly. See the comments with
85 * prepare_range_limit_table (in jdmaster.c) for more info.
88 #define IDCT_range_limit(cinfo) \
89 ((_JSAMPLE *)((cinfo)->sample_range_limit) + _CENTERJSAMPLE)
91 #define RANGE_MASK (_MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
94 /* Extern declarations for the forward and inverse DCT routines. */
96 EXTERN(void) _jpeg_fdct_islow(DCTELEM *data);
97 EXTERN(void) _jpeg_fdct_ifast(DCTELEM *data);
98 EXTERN(void) jpeg_fdct_float(FAST_FLOAT *data);
100 EXTERN(void) _jpeg_idct_islow(j_decompress_ptr cinfo,
101 jpeg_component_info *compptr,
102 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
103 JDIMENSION output_col);
104 EXTERN(void) _jpeg_idct_ifast(j_decompress_ptr cinfo,
105 jpeg_component_info *compptr,
106 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
107 JDIMENSION output_col);
108 EXTERN(void) _jpeg_idct_float(j_decompress_ptr cinfo,
109 jpeg_component_info *compptr,
110 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
111 JDIMENSION output_col);
112 EXTERN(void) _jpeg_idct_7x7(j_decompress_ptr cinfo,
113 jpeg_component_info *compptr, JCOEFPTR coef_block,
114 _JSAMPARRAY output_buf, JDIMENSION output_col);
115 EXTERN(void) _jpeg_idct_6x6(j_decompress_ptr cinfo,
116 jpeg_component_info *compptr, JCOEFPTR coef_block,
117 _JSAMPARRAY output_buf, JDIMENSION output_col);
118 EXTERN(void) _jpeg_idct_5x5(j_decompress_ptr cinfo,
119 jpeg_component_info *compptr, JCOEFPTR coef_block,
120 _JSAMPARRAY output_buf, JDIMENSION output_col);
121 EXTERN(void) _jpeg_idct_4x4(j_decompress_ptr cinfo,
122 jpeg_component_info *compptr, JCOEFPTR coef_block,
123 _JSAMPARRAY output_buf, JDIMENSION output_col);
124 EXTERN(void) _jpeg_idct_3x3(j_decompress_ptr cinfo,
125 jpeg_component_info *compptr, JCOEFPTR coef_block,
126 _JSAMPARRAY output_buf, JDIMENSION output_col);
127 EXTERN(void) _jpeg_idct_2x2(j_decompress_ptr cinfo,
128 jpeg_component_info *compptr, JCOEFPTR coef_block,
129 _JSAMPARRAY output_buf, JDIMENSION output_col);
130 EXTERN(void) _jpeg_idct_1x1(j_decompress_ptr cinfo,
131 jpeg_component_info *compptr, JCOEFPTR coef_block,
132 _JSAMPARRAY output_buf, JDIMENSION output_col);
133 EXTERN(void) _jpeg_idct_9x9(j_decompress_ptr cinfo,
134 jpeg_component_info *compptr, JCOEFPTR coef_block,
135 _JSAMPARRAY output_buf, JDIMENSION output_col);
136 EXTERN(void) _jpeg_idct_10x10(j_decompress_ptr cinfo,
137 jpeg_component_info *compptr,
138 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
139 JDIMENSION output_col);
140 EXTERN(void) _jpeg_idct_11x11(j_decompress_ptr cinfo,
141 jpeg_component_info *compptr,
142 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
143 JDIMENSION output_col);
144 EXTERN(void) _jpeg_idct_12x12(j_decompress_ptr cinfo,
145 jpeg_component_info *compptr,
146 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
147 JDIMENSION output_col);
148 EXTERN(void) _jpeg_idct_13x13(j_decompress_ptr cinfo,
149 jpeg_component_info *compptr,
150 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
151 JDIMENSION output_col);
152 EXTERN(void) _jpeg_idct_14x14(j_decompress_ptr cinfo,
153 jpeg_component_info *compptr,
154 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
155 JDIMENSION output_col);
156 EXTERN(void) _jpeg_idct_15x15(j_decompress_ptr cinfo,
157 jpeg_component_info *compptr,
158 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
159 JDIMENSION output_col);
160 EXTERN(void) _jpeg_idct_16x16(j_decompress_ptr cinfo,
161 jpeg_component_info *compptr,
162 JCOEFPTR coef_block, _JSAMPARRAY output_buf,
163 JDIMENSION output_col);
167 * Macros for handling fixed-point arithmetic; these are used by many
168 * but not all of the DCT/IDCT modules.
170 * All values are expected to be of type JLONG.
171 * Fractional constants are scaled left by CONST_BITS bits.
172 * CONST_BITS is defined within each module using these macros,
173 * and may differ from one module to the next.
176 #define ONE ((JLONG)1)
177 #define CONST_SCALE (ONE << CONST_BITS)
179 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
180 * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
181 * thus causing a lot of useless floating-point operations at run time.
184 #define FIX(x) ((JLONG)((x) * CONST_SCALE + 0.5))
186 /* Descale and correctly round a JLONG value that's scaled by N bits.
187 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
188 * the fudge factor is correct for either sign of X.
191 #define DESCALE(x, n) RIGHT_SHIFT((x) + (ONE << ((n) - 1)), n)
193 /* Multiply a JLONG variable by a JLONG constant to yield a JLONG result.
194 * This macro is used only when the two inputs will actually be no more than
195 * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
196 * full 32x32 multiply. This provides a useful speedup on many machines.
197 * Unfortunately there is no way to specify a 16x16->32 multiply portably
198 * in C, but some C compilers will do the right thing if you provide the
199 * correct combination of casts.
202 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
203 #define MULTIPLY16C16(var, const) (((INT16)(var)) * ((INT16)(const)))
205 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
206 #define MULTIPLY16C16(var, const) (((INT16)(var)) * ((JLONG)(const)))
209 #ifndef MULTIPLY16C16 /* default definition */
210 #define MULTIPLY16C16(var, const) ((var) * (const))
213 /* Same except both inputs are variables. */
215 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
216 #define MULTIPLY16V16(var1, var2) (((INT16)(var1)) * ((INT16)(var2)))
219 #ifndef MULTIPLY16V16 /* default definition */
220 #define MULTIPLY16V16(var1, var2) ((var1) * (var2))