jpeg/jddctmgr.c

   1 /*
   2  * jddctmgr.c
   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 file contains the inverse-DCT management logic.
   9  * This code selects a particular IDCT implementation to be used,
  10  * and it performs related housekeeping chores.  No code in this file
  11  * is executed per IDCT step, only during output pass setup.
  12  *
  13  * Note that the IDCT routines are responsible for performing coefficient
  14  * dequantization as well as the IDCT proper.  This module sets up the
  15  * dequantization multiplier table needed by the IDCT routine.
  16  */
  17
  18 #define JPEG_INTERNALS
  19 #include "jinclude.h"
  20 #include "jpeglib.h"
  21 #include "jdct.h"               /* Private declarations for DCT subsystem */
  22 #ifdef HAVE_SSE2_INTRINSICS
  23 extern int SSE2Available;
  24 #endif
  25
  26 /*
  27  * The decompressor input side (jdinput.c) saves away the appropriate
  28  * quantization table for each component at the start of the first scan
  29  * involving that component.  (This is necessary in order to correctly
  30  * decode files that reuse Q-table slots.)
  31  * When we are ready to make an output pass, the saved Q-table is converted
  32  * to a multiplier table that will actually be used by the IDCT routine.
  33  * The multiplier table contents are IDCT-method-dependent.  To support
  34  * application changes in IDCT method between scans, we can remake the
  35  * multiplier tables if necessary.
  36  * In buffered-image mode, the first output pass may occur before any data
  37  * has been seen for some components, and thus before their Q-tables have
  38  * been saved away.  To handle this case, multiplier tables are preset
  39  * to zeroes; the result of the IDCT will be a neutral gray level.
  40  */
  41
  42
  43 /* Private subobject for this module */
  44
  45 typedef struct {
  46   struct jpeg_inverse_dct pub;  /* public fields */
  47
  48   /* This array contains the IDCT method code that each multiplier table
  49    * is currently set up for, or -1 if it's not yet set up.
  50    * The actual multiplier tables are pointed to by dct_table in the
  51    * per-component comp_info structures.
  52    */
  53   int cur_method[MAX_COMPONENTS];
  54 } my_idct_controller;
  55
  56 typedef my_idct_controller * my_idct_ptr;
  57
  58
  59 /* Allocated multiplier tables: big enough for any supported variant */
  60
  61 typedef union {
  62   ISLOW_MULT_TYPE islow_array[DCTSIZE2];
  63 #ifdef DCT_IFAST_SUPPORTED
  64   IFAST_MULT_TYPE ifast_array[DCTSIZE2];
  65 #endif
  66 #ifdef DCT_FLOAT_SUPPORTED
  67   FLOAT_MULT_TYPE float_array[DCTSIZE2];
  68 #endif
  69 } multiplier_table;
  70
  71
  72 /* The current scaled-IDCT routines require ISLOW-style multiplier tables,
  73  * so be sure to compile that code if either ISLOW or SCALING is requested.
  74  */
  75 #ifdef DCT_ISLOW_SUPPORTED
  76 #define PROVIDE_ISLOW_TABLES
  77 #else
  78 #ifdef IDCT_SCALING_SUPPORTED
  79 #define PROVIDE_ISLOW_TABLES
  80 #endif
  81 #endif
  82
  83 GLOBAL(void)
  84 jpeg_idct_islow_sse2 (
  85         j_decompress_ptr cinfo,
  86         jpeg_component_info * compptr,
  87         JCOEFPTR coef_block,
  88         JSAMPARRAY output_buf,
  89         JDIMENSION output_col);
  90
  91
  92 /*
  93  * Prepare for an output pass.
  94  * Here we select the proper IDCT routine for each component and build
  95  * a matching multiplier table.
  96  */
  97
  98 METHODDEF(void)
  99 start_pass (j_decompress_ptr cinfo)
 100 {
 101   my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
 102   int ci, i;
 103   jpeg_component_info *compptr;
 104   int method = 0;
 105   inverse_DCT_method_ptr method_ptr = NULL;
 106   JQUANT_TBL * qtbl;
 107
 108   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 109        ci++, compptr++) {
 110     /* Select the proper IDCT routine for this component's scaling */
 111     switch (compptr->DCT_scaled_size) {
 112 #ifdef IDCT_SCALING_SUPPORTED
 113     case 1:
 114       method_ptr = jpeg_idct_1x1;
 115       method = JDCT_ISLOW;      /* jidctred uses islow-style table */
 116       break;
 117     case 2:
 118       method_ptr = jpeg_idct_2x2;
 119       method = JDCT_ISLOW;      /* jidctred uses islow-style table */
 120       break;
 121     case 4:
 122       method_ptr = jpeg_idct_4x4;
 123       method = JDCT_ISLOW;      /* jidctred uses islow-style table */
 124       break;
 125 #endif
 126     case DCTSIZE:
 127       switch (cinfo->dct_method) {
 128 #ifdef DCT_ISLOW_SUPPORTED
 129       case JDCT_ISLOW:
 130 #ifdef HAVE_SSE2_INTEL_MNEMONICS
 131                 if(SSE2Available == 1)
 132                 {
 133                         method_ptr = jpeg_idct_islow_sse2;
 134                         method = JDCT_ISLOW;
 135                 }
 136                 else
 137                 {
 138                         method_ptr = jpeg_idct_islow;
 139                         method = JDCT_ISLOW;
 140                 }
 141 #else
 142                 method_ptr = jpeg_idct_islow;
 143                 method = JDCT_ISLOW;
 144
 145 #endif /* HAVE_SSE2_INTEL_MNEMONICS */
 146         break;
 147 #endif
 148 #ifdef DCT_IFAST_SUPPORTED
 149       case JDCT_IFAST:
 150 #ifdef HAVE_SSE2_INTEL_MNEMONICS
 151                 if (SSE2Available==1)
 152                 {
 153                         method_ptr = jpeg_idct_islow_sse2;
 154                         method = JDCT_ISLOW;
 155                 }
 156                 else
 157                 {
 158                         method_ptr = jpeg_idct_ifast;
 159                         method = JDCT_IFAST;
 160                 }
 161 #else
 162                 method_ptr = jpeg_idct_ifast;
 163                 method = JDCT_IFAST;
 164 #endif /* HAVE_SSE2_INTEL_MNEMONICS */
 165         break;
 166
 167 #endif
 168 #ifdef DCT_FLOAT_SUPPORTED
 169       case JDCT_FLOAT:
 170         method_ptr = jpeg_idct_float;
 171         method = JDCT_FLOAT;
 172         break;
 173 #endif
 174       default:
 175         ERREXIT(cinfo, JERR_NOT_COMPILED);
 176         break;
 177       }
 178       break;
 179     default:
 180       ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->DCT_scaled_size);
 181       break;
 182     }
 183     idct->pub.inverse_DCT[ci] = method_ptr;
 184     /* Create multiplier table from quant table.
 185      * However, we can skip this if the component is uninteresting
 186      * or if we already built the table.  Also, if no quant table
 187      * has yet been saved for the component, we leave the
 188      * multiplier table all-zero; we'll be reading zeroes from the
 189      * coefficient controller's buffer anyway.
 190      */
 191     if (! compptr->component_needed || idct->cur_method[ci] == method)
 192       continue;
 193     qtbl = compptr->quant_table;
 194     if (qtbl == NULL)           /* happens if no data yet for component */
 195       continue;
 196     idct->cur_method[ci] = method;
 197     switch (method) {
 198 #ifdef PROVIDE_ISLOW_TABLES
 199     case JDCT_ISLOW:
 200       {
 201         /* For LL&M IDCT method, multipliers are equal to raw quantization
 202          * coefficients, but are stored as ints to ensure access efficiency.
 203          */
 204         ISLOW_MULT_TYPE * ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
 205         for (i = 0; i < DCTSIZE2; i++) {
 206           ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[i];
 207         }
 208       }
 209       break;
 210 #endif
 211 #ifdef DCT_IFAST_SUPPORTED
 212     case JDCT_IFAST:
 213       {
 214         /* For AA&N IDCT method, multipliers are equal to quantization
 215          * coefficients scaled by scalefactor[row]*scalefactor[col], where
 216          *   scalefactor[0] = 1
 217          *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
 218          * For integer operation, the multiplier table is to be scaled by
 219          * IFAST_SCALE_BITS.
 220          */
 221         IFAST_MULT_TYPE * ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
 222 #define CONST_BITS 14
 223         static const INT16 aanscales[DCTSIZE2] = {
 224           /* precomputed values scaled up by 14 bits */
 225           16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
 226           22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
 227           21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
 228           19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
 229           16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
 230           12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
 231            8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
 232            4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
 233         };
 234         SHIFT_TEMPS
 235
 236         for (i = 0; i < DCTSIZE2; i++) {
 237           ifmtbl[i] = (IFAST_MULT_TYPE)
 238             DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
 239                                   (INT32) aanscales[i]),
 240                     CONST_BITS-IFAST_SCALE_BITS);
 241         }
 242       }
 243       break;
 244 #endif
 245 #ifdef DCT_FLOAT_SUPPORTED
 246     case JDCT_FLOAT:
 247       {
 248         /* For float AA&N IDCT method, multipliers are equal to quantization
 249          * coefficients scaled by scalefactor[row]*scalefactor[col], where
 250          *   scalefactor[0] = 1
 251          *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
 252          */
 253         FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
 254         int row, col;
 255         static const double aanscalefactor[DCTSIZE] = {
 256           1.0, 1.387039845, 1.306562965, 1.175875602,
 257           1.0, 0.785694958, 0.541196100, 0.275899379
 258         };
 259
 260         i = 0;
 261         for (row = 0; row < DCTSIZE; row++) {
 262           for (col = 0; col < DCTSIZE; col++) {
 263             fmtbl[i] = (FLOAT_MULT_TYPE)
 264               ((double) qtbl->quantval[i] *
 265                aanscalefactor[row] * aanscalefactor[col]);
 266             i++;
 267           }
 268         }
 269       }
 270       break;
 271 #endif
 272     default:
 273       ERREXIT(cinfo, JERR_NOT_COMPILED);
 274       break;
 275     }
 276   }
 277 }
 278
 279
 280 /*
 281  * Initialize IDCT manager.
 282  */
 283
 284 GLOBAL(void)
 285 jinit_inverse_dct (j_decompress_ptr cinfo)
 286 {
 287   my_idct_ptr idct;
 288   int ci;
 289   jpeg_component_info *compptr;
 290
 291   idct = (my_idct_ptr)
 292     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 293                                 SIZEOF(my_idct_controller));
 294   cinfo->idct = (struct jpeg_inverse_dct *) idct;
 295   idct->pub.start_pass = start_pass;
 296
 297   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 298        ci++, compptr++) {
 299     /* Allocate and pre-zero a multiplier table for each component */
 300     compptr->dct_table =
 301       (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 302                                   SIZEOF(multiplier_table));
 303     MEMZERO(compptr->dct_table, SIZEOF(multiplier_table));
 304     /* Mark multiplier table not yet set up for any method */
 305     idct->cur_method[ci] = -1;
 306   }
 307 }