media/libjpeg/jdct.h@b8a032363ba2 (annotated)
media/libjpeg/jdct.h
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
| michael@0 | 1 | /* |
| michael@0 | 2 | * jdct.h |
| michael@0 | 3 | * |
| michael@0 | 4 | * Copyright (C) 1994-1996, Thomas G. Lane. |
| michael@0 | 5 | * This file is part of the Independent JPEG Group's software. |
| michael@0 | 6 | * For conditions of distribution and use, see the accompanying README file. |
| michael@0 | 7 | * |
| michael@0 | 8 | * This include file contains common declarations for the forward and |
| michael@0 | 9 | * inverse DCT modules. These declarations are private to the DCT managers |
| michael@0 | 10 | * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms. |
| michael@0 | 11 | * The individual DCT algorithms are kept in separate files to ease |
| michael@0 | 12 | * machine-dependent tuning (e.g., assembly coding). |
| michael@0 | 13 | */ |
| michael@0 | 14 | |
| michael@0 | 15 | |
| michael@0 | 16 | /* |
| michael@0 | 17 | * A forward DCT routine is given a pointer to a work area of type DCTELEM[]; |
| michael@0 | 18 | * the DCT is to be performed in-place in that buffer. Type DCTELEM is int |
| michael@0 | 19 | * for 8-bit samples, INT32 for 12-bit samples. (NOTE: Floating-point DCT |
| michael@0 | 20 | * implementations use an array of type FAST_FLOAT, instead.) |
| michael@0 | 21 | * The DCT inputs are expected to be signed (range +-CENTERJSAMPLE). |
| michael@0 | 22 | * The DCT outputs are returned scaled up by a factor of 8; they therefore |
| michael@0 | 23 | * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This |
| michael@0 | 24 | * convention improves accuracy in integer implementations and saves some |
| michael@0 | 25 | * work in floating-point ones. |
| michael@0 | 26 | * Quantization of the output coefficients is done by jcdctmgr.c. This |
| michael@0 | 27 | * step requires an unsigned type and also one with twice the bits. |
| michael@0 | 28 | */ |
| michael@0 | 29 | |
| michael@0 | 30 | #if BITS_IN_JSAMPLE == 8 |
| michael@0 | 31 | #ifndef WITH_SIMD |
| michael@0 | 32 | typedef int DCTELEM; /* 16 or 32 bits is fine */ |
| michael@0 | 33 | typedef unsigned int UDCTELEM; |
| michael@0 | 34 | typedef unsigned long long UDCTELEM2; |
| michael@0 | 35 | #else |
| michael@0 | 36 | typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */ |
| michael@0 | 37 | typedef unsigned short UDCTELEM; |
| michael@0 | 38 | typedef unsigned int UDCTELEM2; |
| michael@0 | 39 | #endif |
| michael@0 | 40 | #else |
| michael@0 | 41 | typedef INT32 DCTELEM; /* must have 32 bits */ |
| michael@0 | 42 | typedef UINT32 UDCTELEM; |
| michael@0 | 43 | typedef unsigned long long UDCTELEM2; |
| michael@0 | 44 | #endif |
| michael@0 | 45 | |
| michael@0 | 46 | |
| michael@0 | 47 | /* |
| michael@0 | 48 | * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer |
| michael@0 | 49 | * to an output sample array. The routine must dequantize the input data as |
| michael@0 | 50 | * well as perform the IDCT; for dequantization, it uses the multiplier table |
| michael@0 | 51 | * pointed to by compptr->dct_table. The output data is to be placed into the |
| michael@0 | 52 | * sample array starting at a specified column. (Any row offset needed will |
| michael@0 | 53 | * be applied to the array pointer before it is passed to the IDCT code.) |
| michael@0 | 54 | * Note that the number of samples emitted by the IDCT routine is |
| michael@0 | 55 | * DCT_scaled_size * DCT_scaled_size. |
| michael@0 | 56 | */ |
| michael@0 | 57 | |
| michael@0 | 58 | /* typedef inverse_DCT_method_ptr is declared in jpegint.h */ |
| michael@0 | 59 | |
| michael@0 | 60 | /* |
| michael@0 | 61 | * Each IDCT routine has its own ideas about the best dct_table element type. |
| michael@0 | 62 | */ |
| michael@0 | 63 | |
| michael@0 | 64 | typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */ |
| michael@0 | 65 | #if BITS_IN_JSAMPLE == 8 |
| michael@0 | 66 | typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */ |
| michael@0 | 67 | #define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */ |
| michael@0 | 68 | #else |
| michael@0 | 69 | typedef INT32 IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */ |
| michael@0 | 70 | #define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */ |
| michael@0 | 71 | #endif |
| michael@0 | 72 | typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */ |
| michael@0 | 73 | |
| michael@0 | 74 | |
| michael@0 | 75 | /* |
| michael@0 | 76 | * Each IDCT routine is responsible for range-limiting its results and |
| michael@0 | 77 | * converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could |
| michael@0 | 78 | * be quite far out of range if the input data is corrupt, so a bulletproof |
| michael@0 | 79 | * range-limiting step is required. We use a mask-and-table-lookup method |
| michael@0 | 80 | * to do the combined operations quickly. See the comments with |
| michael@0 | 81 | * prepare_range_limit_table (in jdmaster.c) for more info. |
| michael@0 | 82 | */ |
| michael@0 | 83 | |
| michael@0 | 84 | #define IDCT_range_limit(cinfo) ((cinfo)->sample_range_limit + CENTERJSAMPLE) |
| michael@0 | 85 | |
| michael@0 | 86 | #define RANGE_MASK (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */ |
| michael@0 | 87 | |
| michael@0 | 88 | |
| michael@0 | 89 | /* Short forms of external names for systems with brain-damaged linkers. */ |
| michael@0 | 90 | |
| michael@0 | 91 | #ifdef NEED_SHORT_EXTERNAL_NAMES |
| michael@0 | 92 | #define jpeg_fdct_islow jFDislow |
| michael@0 | 93 | #define jpeg_fdct_ifast jFDifast |
| michael@0 | 94 | #define jpeg_fdct_float jFDfloat |
| michael@0 | 95 | #define jpeg_idct_islow jRDislow |
| michael@0 | 96 | #define jpeg_idct_ifast jRDifast |
| michael@0 | 97 | #define jpeg_idct_float jRDfloat |
| michael@0 | 98 | #define jpeg_idct_7x7 jRD7x7 |
| michael@0 | 99 | #define jpeg_idct_6x6 jRD6x6 |
| michael@0 | 100 | #define jpeg_idct_5x5 jRD5x5 |
| michael@0 | 101 | #define jpeg_idct_4x4 jRD4x4 |
| michael@0 | 102 | #define jpeg_idct_3x3 jRD3x3 |
| michael@0 | 103 | #define jpeg_idct_2x2 jRD2x2 |
| michael@0 | 104 | #define jpeg_idct_1x1 jRD1x1 |
| michael@0 | 105 | #define jpeg_idct_9x9 jRD9x9 |
| michael@0 | 106 | #define jpeg_idct_10x10 jRD10x10 |
| michael@0 | 107 | #define jpeg_idct_11x11 jRD11x11 |
| michael@0 | 108 | #define jpeg_idct_12x12 jRD12x12 |
| michael@0 | 109 | #define jpeg_idct_13x13 jRD13x13 |
| michael@0 | 110 | #define jpeg_idct_14x14 jRD14x14 |
| michael@0 | 111 | #define jpeg_idct_15x15 jRD15x15 |
| michael@0 | 112 | #define jpeg_idct_16x16 jRD16x16 |
| michael@0 | 113 | #endif /* NEED_SHORT_EXTERNAL_NAMES */ |
| michael@0 | 114 | |
| michael@0 | 115 | /* Extern declarations for the forward and inverse DCT routines. */ |
| michael@0 | 116 | |
| michael@0 | 117 | EXTERN(void) jpeg_fdct_islow JPP((DCTELEM * data)); |
| michael@0 | 118 | EXTERN(void) jpeg_fdct_ifast JPP((DCTELEM * data)); |
| michael@0 | 119 | EXTERN(void) jpeg_fdct_float JPP((FAST_FLOAT * data)); |
| michael@0 | 120 | |
| michael@0 | 121 | EXTERN(void) jpeg_idct_islow |
| michael@0 | 122 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 123 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 124 | EXTERN(void) jpeg_idct_ifast |
| michael@0 | 125 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 126 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 127 | EXTERN(void) jpeg_idct_float |
| michael@0 | 128 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 129 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 130 | EXTERN(void) jpeg_idct_7x7 |
| michael@0 | 131 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 132 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 133 | EXTERN(void) jpeg_idct_6x6 |
| michael@0 | 134 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 135 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 136 | EXTERN(void) jpeg_idct_5x5 |
| michael@0 | 137 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 138 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 139 | EXTERN(void) jpeg_idct_4x4 |
| michael@0 | 140 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 141 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 142 | EXTERN(void) jpeg_idct_3x3 |
| michael@0 | 143 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 144 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 145 | EXTERN(void) jpeg_idct_2x2 |
| michael@0 | 146 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 147 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 148 | EXTERN(void) jpeg_idct_1x1 |
| michael@0 | 149 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 150 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 151 | EXTERN(void) jpeg_idct_9x9 |
| michael@0 | 152 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 153 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 154 | EXTERN(void) jpeg_idct_10x10 |
| michael@0 | 155 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 156 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 157 | EXTERN(void) jpeg_idct_11x11 |
| michael@0 | 158 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 159 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 160 | EXTERN(void) jpeg_idct_12x12 |
| michael@0 | 161 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 162 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 163 | EXTERN(void) jpeg_idct_13x13 |
| michael@0 | 164 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 165 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 166 | EXTERN(void) jpeg_idct_14x14 |
| michael@0 | 167 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 168 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 169 | EXTERN(void) jpeg_idct_15x15 |
| michael@0 | 170 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 171 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 172 | EXTERN(void) jpeg_idct_16x16 |
| michael@0 | 173 | JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr, |
| michael@0 | 174 | JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col)); |
| michael@0 | 175 | |
| michael@0 | 176 | |
| michael@0 | 177 | /* |
| michael@0 | 178 | * Macros for handling fixed-point arithmetic; these are used by many |
| michael@0 | 179 | * but not all of the DCT/IDCT modules. |
| michael@0 | 180 | * |
| michael@0 | 181 | * All values are expected to be of type INT32. |
| michael@0 | 182 | * Fractional constants are scaled left by CONST_BITS bits. |
| michael@0 | 183 | * CONST_BITS is defined within each module using these macros, |
| michael@0 | 184 | * and may differ from one module to the next. |
| michael@0 | 185 | */ |
| michael@0 | 186 | |
| michael@0 | 187 | #define ONE ((INT32) 1) |
| michael@0 | 188 | #define CONST_SCALE (ONE << CONST_BITS) |
| michael@0 | 189 | |
| michael@0 | 190 | /* Convert a positive real constant to an integer scaled by CONST_SCALE. |
| michael@0 | 191 | * Caution: some C compilers fail to reduce "FIX(constant)" at compile time, |
| michael@0 | 192 | * thus causing a lot of useless floating-point operations at run time. |
| michael@0 | 193 | */ |
| michael@0 | 194 | |
| michael@0 | 195 | #define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5)) |
| michael@0 | 196 | |
| michael@0 | 197 | /* Descale and correctly round an INT32 value that's scaled by N bits. |
| michael@0 | 198 | * We assume RIGHT_SHIFT rounds towards minus infinity, so adding |
| michael@0 | 199 | * the fudge factor is correct for either sign of X. |
| michael@0 | 200 | */ |
| michael@0 | 201 | |
| michael@0 | 202 | #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n) |
| michael@0 | 203 | |
| michael@0 | 204 | /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. |
| michael@0 | 205 | * This macro is used only when the two inputs will actually be no more than |
| michael@0 | 206 | * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a |
| michael@0 | 207 | * full 32x32 multiply. This provides a useful speedup on many machines. |
| michael@0 | 208 | * Unfortunately there is no way to specify a 16x16->32 multiply portably |
| michael@0 | 209 | * in C, but some C compilers will do the right thing if you provide the |
| michael@0 | 210 | * correct combination of casts. |
| michael@0 | 211 | */ |
| michael@0 | 212 | |
| michael@0 | 213 | #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ |
| michael@0 | 214 | #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT16) (const))) |
| michael@0 | 215 | #endif |
| michael@0 | 216 | #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */ |
| michael@0 | 217 | #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT32) (const))) |
| michael@0 | 218 | #endif |
| michael@0 | 219 | |
| michael@0 | 220 | #ifndef MULTIPLY16C16 /* default definition */ |
| michael@0 | 221 | #define MULTIPLY16C16(var,const) ((var) * (const)) |
| michael@0 | 222 | #endif |
| michael@0 | 223 | |
| michael@0 | 224 | /* Same except both inputs are variables. */ |
| michael@0 | 225 | |
| michael@0 | 226 | #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ |
| michael@0 | 227 | #define MULTIPLY16V16(var1,var2) (((INT16) (var1)) * ((INT16) (var2))) |
| michael@0 | 228 | #endif |
| michael@0 | 229 | |
| michael@0 | 230 | #ifndef MULTIPLY16V16 /* default definition */ |
| michael@0 | 231 | #define MULTIPLY16V16(var1,var2) ((var1) * (var2)) |
| michael@0 | 232 | #endif |
