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. This 27 * step requires an unsigned type and also one with twice the bits. 28 */ 29 30 #if BITS_IN_JSAMPLE == 8 31 #ifndef WITH_SIMD 32 typedef int DCTELEM; /* 16 or 32 bits is fine */ 33 typedef unsigned int UDCTELEM; 34 typedef unsigned long long UDCTELEM2; 35 #else 36 typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */ 37 typedef unsigned short UDCTELEM; 38 typedef unsigned int UDCTELEM2; 39 #endif 40 #else 41 typedef INT32 DCTELEM; /* must have 32 bits */ 42 typedef UINT32 UDCTELEM; 43 typedef unsigned long long UDCTELEM2; 44 #endif 45 46 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 64 typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */ 65 #if BITS_IN_JSAMPLE == 8 66 typedef 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 69 typedef INT32 IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */ 70 #define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */ 71 #endif 72 typedef 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_7x7 jRD7x7 99 #define jpeg_idct_6x6 jRD6x6 100 #define jpeg_idct_5x5 jRD5x5 101 #define jpeg_idct_4x4 jRD4x4 102 #define jpeg_idct_3x3 jRD3x3 103 #define jpeg_idct_2x2 jRD2x2 104 #define jpeg_idct_1x1 jRD1x1 105 #define jpeg_idct_9x9 jRD9x9 106 #define jpeg_idct_10x10 jRD10x10 107 #define jpeg_idct_11x11 jRD11x11 108 #define jpeg_idct_12x12 jRD12x12 109 #define jpeg_idct_13x13 jRD13x13 110 #define jpeg_idct_14x14 jRD14x14 111 #define jpeg_idct_15x15 jRD15x15 112 #define jpeg_idct_16x16 jRD16x16 113 #endif /* NEED_SHORT_EXTERNAL_NAMES */ 114 115 /* Extern declarations for the forward and inverse DCT routines. */ 116 117 EXTERN(void) jpeg_fdct_islow JPP((DCTELEM * data)); 118 EXTERN(void) jpeg_fdct_ifast JPP((DCTELEM * data)); 119 EXTERN(void) jpeg_fdct_float JPP((FAST_FLOAT * data)); 120 121 EXTERN(void) jpeg_idct_islow 122 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 123 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 124 EXTERN(void) jpeg_idct_ifast 125 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 126 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 127 EXTERN(void) jpeg_idct_float 128 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 129 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 130 EXTERN(void) jpeg_idct_7x7 131 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 132 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 133 EXTERN(void) jpeg_idct_6x6 134 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 135 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 136 EXTERN(void) jpeg_idct_5x5 137 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 138 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 139 EXTERN(void) jpeg_idct_4x4 140 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 141 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 142 EXTERN(void) jpeg_idct_3x3 143 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 144 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 145 EXTERN(void) jpeg_idct_2x2 146 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 147 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 148 EXTERN(void) jpeg_idct_1x1 149 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 150 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 151 EXTERN(void) jpeg_idct_9x9 152 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 153 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 154 EXTERN(void) jpeg_idct_10x10 155 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 156 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 157 EXTERN(void) jpeg_idct_11x11 158 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 159 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 160 EXTERN(void) jpeg_idct_12x12 161 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 162 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 163 EXTERN(void) jpeg_idct_13x13 164 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 165 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 166 EXTERN(void) jpeg_idct_14x14 167 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 168 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 169 EXTERN(void) jpeg_idct_15x15 170 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 171 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 172 EXTERN(void) jpeg_idct_16x16 173 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, 174 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); 175 176 177 /* 178 * Macros for handling fixed-point arithmetic; these are used by many 179 * but not all of the DCT/IDCT modules. 180 * 181 * All values are expected to be of type INT32. 182 * Fractional constants are scaled left by CONST_BITS bits. 183 * CONST_BITS is defined within each module using these macros, 184 * and may differ from one module to the next. 185 */ 186 187 #define ONE ((INT32) 1) 188 #define CONST_SCALE (ONE << CONST_BITS) 189 190 /* Convert a positive real constant to an integer scaled by CONST_SCALE. 191 * Caution: some C compilers fail to reduce "FIX(constant)" at compile time, 192 * thus causing a lot of useless floating-point operations at run time. 193 */ 194 195 #define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5)) 196 197 /* Descale and correctly round an INT32 value that's scaled by N bits. 198 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding 199 * the fudge factor is correct for either sign of X. 200 */ 201 202 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n) 203 204 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. 205 * This macro is used only when the two inputs will actually be no more than 206 * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a 207 * full 32x32 multiply. This provides a useful speedup on many machines. 208 * Unfortunately there is no way to specify a 16x16->32 multiply portably 209 * in C, but some C compilers will do the right thing if you provide the 210 * correct combination of casts. 211 */ 212 213 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ 214 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT16) (const))) 215 #endif 216 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */ 217 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT32) (const))) 218 #endif 219 220 #ifndef MULTIPLY16C16 /* default definition */ 221 #define MULTIPLY16C16(var,const) ((var) * (const)) 222 #endif 223 224 /* Same except both inputs are variables. */ 225 226 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ 227 #define MULTIPLY16V16(var1,var2) (((INT16) (var1)) * ((INT16) (var2))) 228 #endif 229 230 #ifndef MULTIPLY16V16 /* default definition */ 231 #define MULTIPLY16V16(var1,var2) ((var1) * (var2)) 232 #endif 233