1 /* 2 * jdct.h 3 * 4 * This file was part of the Independent JPEG Group's software: 5 * Copyright (C) 1994-1996, Thomas G. Lane. 6 * It was modified by The libjpeg-turbo Project to include only code relevant 7 * to libjpeg-turbo. 8 * For conditions of distribution and use, see the accompanying README file. 9 * 10 * This include file contains common declarations for the forward and 11 * inverse DCT modules. These declarations are private to the DCT managers 12 * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms. 13 * The individual DCT algorithms are kept in separate files to ease 14 * machine-dependent tuning (e.g., assembly coding). 15 */ 16 17 18 /* 19 * A forward DCT routine is given a pointer to a work area of type DCTELEM[]; 20 * the DCT is to be performed in-place in that buffer. Type DCTELEM is int 21 * for 8-bit samples, INT32 for 12-bit samples. (NOTE: Floating-point DCT 22 * implementations use an array of type FAST_FLOAT, instead.) 23 * The DCT inputs are expected to be signed (range +-CENTERJSAMPLE). 24 * The DCT outputs are returned scaled up by a factor of 8; they therefore 25 * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This 26 * convention improves accuracy in integer implementations and saves some 27 * work in floating-point ones. 28 * Quantization of the output coefficients is done by jcdctmgr.c. This 29 * step requires an unsigned type and also one with twice the bits. 30 */ 31 32 #if BITS_IN_JSAMPLE == 8 33 #ifndef WITH_SIMD 34 typedef int DCTELEM; /* 16 or 32 bits is fine */ 35 typedef unsigned int UDCTELEM; 36 typedef unsigned long long UDCTELEM2; 37 #else 38 typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */ 39 typedef unsigned short UDCTELEM; 40 typedef unsigned int UDCTELEM2; 41 #endif 42 #else 43 typedef INT32 DCTELEM; /* must have 32 bits */ 44 typedef unsigned long long UDCTELEM2; 45 #endif 46 47 48 /* 49 * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer 50 * to an output sample array. The routine must dequantize the input data as 51 * well as perform the IDCT; for dequantization, it uses the multiplier table 52 * pointed to by compptr->dct_table. The output data is to be placed into the 53 * sample array starting at a specified column. (Any row offset needed will 54 * be applied to the array pointer before it is passed to the IDCT code.) 55 * Note that the number of samples emitted by the IDCT routine is 56 * DCT_scaled_size * DCT_scaled_size. 57 */ 58 59 /* typedef inverse_DCT_method_ptr is declared in jpegint.h */ 60 61 /* 62 * Each IDCT routine has its own ideas about the best dct_table element type. 63 */ 64 65 typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */ 66 #if BITS_IN_JSAMPLE == 8 67 typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */ 68 #define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */ 69 #else 70 typedef INT32 IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */ 71 #define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */ 72 #endif 73 typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */ 74 75 76 /* 77 * Each IDCT routine is responsible for range-limiting its results and 78 * converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could 79 * be quite far out of range if the input data is corrupt, so a bulletproof 80 * range-limiting step is required. We use a mask-and-table-lookup method 81 * to do the combined operations quickly. See the comments with 82 * prepare_range_limit_table (in jdmaster.c) for more info. 83 */ 84 85 #define IDCT_range_limit(cinfo) ((cinfo)->sample_range_limit + CENTERJSAMPLE) 86 87 #define RANGE_MASK (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */ 88 89 90 /* Extern declarations for the forward and inverse DCT routines. */ 91 92 EXTERN(void) jpeg_fdct_islow (DCTELEM * data); 93 EXTERN(void) jpeg_fdct_ifast (DCTELEM * data); 94 EXTERN(void) jpeg_fdct_float (FAST_FLOAT * data); 95 96 EXTERN(void) jpeg_idct_islow 97 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 98 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 99 EXTERN(void) jpeg_idct_ifast 100 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 101 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 102 EXTERN(void) jpeg_idct_float 103 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 104 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 105 EXTERN(void) jpeg_idct_7x7 106 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 107 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 108 EXTERN(void) jpeg_idct_6x6 109 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 110 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 111 EXTERN(void) jpeg_idct_5x5 112 (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 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 116 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 117 EXTERN(void) jpeg_idct_3x3 118 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 119 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 120 EXTERN(void) jpeg_idct_2x2 121 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 122 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 123 EXTERN(void) jpeg_idct_1x1 124 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 125 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 126 EXTERN(void) jpeg_idct_9x9 127 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 128 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 129 EXTERN(void) jpeg_idct_10x10 130 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 131 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 132 EXTERN(void) jpeg_idct_11x11 133 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 134 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 135 EXTERN(void) jpeg_idct_12x12 136 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 137 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 138 EXTERN(void) jpeg_idct_13x13 139 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 140 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 141 EXTERN(void) jpeg_idct_14x14 142 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 143 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 144 EXTERN(void) jpeg_idct_15x15 145 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 146 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 147 EXTERN(void) jpeg_idct_16x16 148 (j_decompress_ptr cinfo, jpeg_component_info * compptr, 149 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col); 150 151 152 /* 153 * Macros for handling fixed-point arithmetic; these are used by many 154 * but not all of the DCT/IDCT modules. 155 * 156 * All values are expected to be of type INT32. 157 * Fractional constants are scaled left by CONST_BITS bits. 158 * CONST_BITS is defined within each module using these macros, 159 * and may differ from one module to the next. 160 */ 161 162 #define ONE ((INT32) 1) 163 #define CONST_SCALE (ONE << CONST_BITS) 164 165 /* Convert a positive real constant to an integer scaled by CONST_SCALE. 166 * Caution: some C compilers fail to reduce "FIX(constant)" at compile time, 167 * thus causing a lot of useless floating-point operations at run time. 168 */ 169 170 #define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5)) 171 172 /* Descale and correctly round an INT32 value that's scaled by N bits. 173 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding 174 * the fudge factor is correct for either sign of X. 175 */ 176 177 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n) 178 179 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. 180 * This macro is used only when the two inputs will actually be no more than 181 * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a 182 * full 32x32 multiply. This provides a useful speedup on many machines. 183 * Unfortunately there is no way to specify a 16x16->32 multiply portably 184 * in C, but some C compilers will do the right thing if you provide the 185 * correct combination of casts. 186 */ 187 188 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ 189 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT16) (const))) 190 #endif 191 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */ 192 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT32) (const))) 193 #endif 194 195 #ifndef MULTIPLY16C16 /* default definition */ 196 #define MULTIPLY16C16(var,const) ((var) * (const)) 197 #endif 198 199 /* Same except both inputs are variables. */ 200 201 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ 202 #define MULTIPLY16V16(var1,var2) (((INT16) (var1)) * ((INT16) (var2))) 203 #endif 204 205 #ifndef MULTIPLY16V16 /* default definition */ 206 #define MULTIPLY16V16(var1,var2) ((var1) * (var2)) 207 #endif 208