DOOM-3-BFG

DOOM 3 BFG Edition
Log | Files | Refs

jidctred.cpp (15072B)


      1 /*
      2  * jidctred.c
      3  *
      4  * Copyright (C) 1994, 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 inverse-DCT routines that produce reduced-size output:
      9  * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.
     10  *
     11  * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M)
     12  * algorithm used in jidctint.c.  We simply replace each 8-to-8 1-D IDCT step
     13  * with an 8-to-4 step that produces the four averages of two adjacent outputs
     14  * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output).
     15  * These steps were derived by computing the corresponding values at the end
     16  * of the normal LL&M code, then simplifying as much as possible.
     17  *
     18  * 1x1 is trivial: just take the DC coefficient divided by 8.
     19  *
     20  * See jidctint.c for additional comments.
     21  */
     22 
     23 #define JPEG_INTERNALS
     24 #include "jinclude.h"
     25 #include "jpeglib.h"
     26 #include "jdct.h"        /* Private declarations for DCT subsystem */
     27 
     28 #ifdef IDCT_SCALING_SUPPORTED
     29 
     30 
     31 /*
     32  * This module is specialized to the case DCTSIZE = 8.
     33  */
     34 
     35 #if DCTSIZE != 8
     36 Sorry, this code only copes with 8 x8 DCTs.  /* deliberate syntax err */
     37     #endif
     38 
     39 
     40 /* Scaling is the same as in jidctint.c. */
     41 
     42 #if BITS_IN_JSAMPLE == 8
     43 #define CONST_BITS  13
     44 #define PASS1_BITS  2
     45 #else
     46 #define CONST_BITS  13
     47 #define PASS1_BITS  1       /* lose a little precision to avoid overflow */
     48 #endif
     49 
     50 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
     51  * causing a lot of useless floating-point operations at run time.
     52  * To get around this we use the following pre-calculated constants.
     53  * If you change CONST_BITS you may want to add appropriate values.
     54  * (With a reasonable C compiler, you can just rely on the FIX() macro...)
     55  */
     56 
     57 #if CONST_BITS == 13
     58 #define FIX_0_211164243  ( (INT32)  1730 )    /* FIX(0.211164243) */
     59 #define FIX_0_509795579  ( (INT32)  4176 )    /* FIX(0.509795579) */
     60 #define FIX_0_601344887  ( (INT32)  4926 )    /* FIX(0.601344887) */
     61 #define FIX_0_720959822  ( (INT32)  5906 )    /* FIX(0.720959822) */
     62 #define FIX_0_765366865  ( (INT32)  6270 )    /* FIX(0.765366865) */
     63 #define FIX_0_850430095  ( (INT32)  6967 )    /* FIX(0.850430095) */
     64 #define FIX_0_899976223  ( (INT32)  7373 )    /* FIX(0.899976223) */
     65 #define FIX_1_061594337  ( (INT32)  8697 )    /* FIX(1.061594337) */
     66 #define FIX_1_272758580  ( (INT32)  10426 )   /* FIX(1.272758580) */
     67 #define FIX_1_451774981  ( (INT32)  11893 )   /* FIX(1.451774981) */
     68 #define FIX_1_847759065  ( (INT32)  15137 )   /* FIX(1.847759065) */
     69 #define FIX_2_172734803  ( (INT32)  17799 )   /* FIX(2.172734803) */
     70 #define FIX_2_562915447  ( (INT32)  20995 )   /* FIX(2.562915447) */
     71 #define FIX_3_624509785  ( (INT32)  29692 )   /* FIX(3.624509785) */
     72 #else
     73 #define FIX_0_211164243  FIX( 0.211164243 )
     74 #define FIX_0_509795579  FIX( 0.509795579 )
     75 #define FIX_0_601344887  FIX( 0.601344887 )
     76 #define FIX_0_720959822  FIX( 0.720959822 )
     77 #define FIX_0_765366865  FIX( 0.765366865 )
     78 #define FIX_0_850430095  FIX( 0.850430095 )
     79 #define FIX_0_899976223  FIX( 0.899976223 )
     80 #define FIX_1_061594337  FIX( 1.061594337 )
     81 #define FIX_1_272758580  FIX( 1.272758580 )
     82 #define FIX_1_451774981  FIX( 1.451774981 )
     83 #define FIX_1_847759065  FIX( 1.847759065 )
     84 #define FIX_2_172734803  FIX( 2.172734803 )
     85 #define FIX_2_562915447  FIX( 2.562915447 )
     86 #define FIX_3_624509785  FIX( 3.624509785 )
     87 #endif
     88 
     89 
     90 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
     91  * For 8-bit samples with the recommended scaling, all the variable
     92  * and constant values involved are no more than 16 bits wide, so a
     93  * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
     94  * For 12-bit samples, a full 32-bit multiplication will be needed.
     95  */
     96 
     97 #if BITS_IN_JSAMPLE == 8
     98 #define MULTIPLY( var, const )  MULTIPLY16C16( var, const )
     99 #else
    100 #define MULTIPLY( var, const )  ( ( var ) * ( const ) )
    101 #endif
    102 
    103 
    104 /* Dequantize a coefficient by multiplying it by the multiplier-table
    105  * entry; produce an int result.  In this module, both inputs and result
    106  * are 16 bits or less, so either int or short multiply will work.
    107  */
    108 
    109 #define DEQUANTIZE( coef, quantval )  ( ( (ISLOW_MULT_TYPE) ( coef ) ) * ( quantval ) )
    110 
    111 
    112 /*
    113  * Perform dequantization and inverse DCT on one block of coefficients,
    114  * producing a reduced-size 4x4 output block.
    115  */
    116 
    117 GLOBAL void
    118 jpeg_idct_4x4( j_decompress_ptr cinfo, jpeg_component_info * compptr,
    119                JCOEFPTR coef_block,
    120                JSAMPARRAY output_buf, JDIMENSION output_col ) {
    121     INT32 tmp0, tmp2, tmp10, tmp12;
    122     INT32 z1, z2, z3, z4;
    123     JCOEFPTR inptr;
    124     ISLOW_MULT_TYPE * quantptr;
    125     int * wsptr;
    126     JSAMPROW outptr;
    127     JSAMPLE * range_limit = IDCT_range_limit( cinfo );
    128     int ctr;
    129     int workspace[DCTSIZE * 4];/* buffers data between passes */
    130     SHIFT_TEMPS
    131 
    132     /* Pass 1: process columns from input, store into work array. */
    133 
    134     inptr = coef_block;
    135     quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
    136     wsptr = workspace;
    137     for ( ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr-- ) {
    138         /* Don't bother to process column 4, because second pass won't use it */
    139         if ( ctr == DCTSIZE - 4 ) {
    140             continue;
    141         }
    142         if ( ( inptr[DCTSIZE * 1] | inptr[DCTSIZE * 2] | inptr[DCTSIZE * 3] |
    143                inptr[DCTSIZE * 5] | inptr[DCTSIZE * 6] | inptr[DCTSIZE * 7] ) == 0 ) {
    144             /* AC terms all zero; we need not examine term 4 for 4x4 output */
    145             int dcval = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] ) << PASS1_BITS;
    146 
    147             wsptr[DCTSIZE * 0] = dcval;
    148             wsptr[DCTSIZE * 1] = dcval;
    149             wsptr[DCTSIZE * 2] = dcval;
    150             wsptr[DCTSIZE * 3] = dcval;
    151 
    152             continue;
    153         }
    154 
    155         /* Even part */
    156 
    157         tmp0 = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] );
    158         tmp0 <<= ( CONST_BITS + 1 );
    159 
    160         z2 = DEQUANTIZE( inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2] );
    161         z3 = DEQUANTIZE( inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6] );
    162 
    163         tmp2 = MULTIPLY( z2, FIX_1_847759065 ) + MULTIPLY( z3, -FIX_0_765366865 );
    164 
    165         tmp10 = tmp0 + tmp2;
    166         tmp12 = tmp0 - tmp2;
    167 
    168         /* Odd part */
    169 
    170         z1 = DEQUANTIZE( inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7] );
    171         z2 = DEQUANTIZE( inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5] );
    172         z3 = DEQUANTIZE( inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3] );
    173         z4 = DEQUANTIZE( inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1] );
    174 
    175         tmp0 = MULTIPLY( z1, -FIX_0_211164243 )/* sqrt(2) * (c3-c1) */
    176                + MULTIPLY( z2, FIX_1_451774981 )/* sqrt(2) * (c3+c7) */
    177                + MULTIPLY( z3, -FIX_2_172734803 )/* sqrt(2) * (-c1-c5) */
    178                + MULTIPLY( z4, FIX_1_061594337 );/* sqrt(2) * (c5+c7) */
    179 
    180         tmp2 = MULTIPLY( z1, -FIX_0_509795579 )/* sqrt(2) * (c7-c5) */
    181                + MULTIPLY( z2, -FIX_0_601344887 )/* sqrt(2) * (c5-c1) */
    182                + MULTIPLY( z3, FIX_0_899976223 )/* sqrt(2) * (c3-c7) */
    183                + MULTIPLY( z4, FIX_2_562915447 );/* sqrt(2) * (c1+c3) */
    184 
    185         /* Final output stage */
    186 
    187         wsptr[DCTSIZE * 0] = (int) DESCALE( tmp10 + tmp2, CONST_BITS - PASS1_BITS + 1 );
    188         wsptr[DCTSIZE * 3] = (int) DESCALE( tmp10 - tmp2, CONST_BITS - PASS1_BITS + 1 );
    189         wsptr[DCTSIZE * 1] = (int) DESCALE( tmp12 + tmp0, CONST_BITS - PASS1_BITS + 1 );
    190         wsptr[DCTSIZE * 2] = (int) DESCALE( tmp12 - tmp0, CONST_BITS - PASS1_BITS + 1 );
    191     }
    192 
    193     /* Pass 2: process 4 rows from work array, store into output array. */
    194 
    195     wsptr = workspace;
    196     for ( ctr = 0; ctr < 4; ctr++ ) {
    197         outptr = output_buf[ctr] + output_col;
    198         /* It's not clear whether a zero row test is worthwhile here ... */
    199 
    200 #ifndef NO_ZERO_ROW_TEST
    201         if ( ( wsptr[1] | wsptr[2] | wsptr[3] | wsptr[5] | wsptr[6] |
    202                wsptr[7] ) == 0 ) {
    203             /* AC terms all zero */
    204             JSAMPLE dcval = range_limit[(int) DESCALE( (INT32) wsptr[0], PASS1_BITS + 3 )
    205                                         & RANGE_MASK];
    206 
    207             outptr[0] = dcval;
    208             outptr[1] = dcval;
    209             outptr[2] = dcval;
    210             outptr[3] = dcval;
    211 
    212             wsptr += DCTSIZE;/* advance pointer to next row */
    213             continue;
    214         }
    215 #endif
    216 
    217         /* Even part */
    218 
    219         tmp0 = ( (INT32) wsptr[0] ) << ( CONST_BITS + 1 );
    220 
    221         tmp2 = MULTIPLY( (INT32) wsptr[2], FIX_1_847759065 )
    222                + MULTIPLY( (INT32) wsptr[6], -FIX_0_765366865 );
    223 
    224         tmp10 = tmp0 + tmp2;
    225         tmp12 = tmp0 - tmp2;
    226 
    227         /* Odd part */
    228 
    229         z1 = (INT32) wsptr[7];
    230         z2 = (INT32) wsptr[5];
    231         z3 = (INT32) wsptr[3];
    232         z4 = (INT32) wsptr[1];
    233 
    234         tmp0 = MULTIPLY( z1, -FIX_0_211164243 )/* sqrt(2) * (c3-c1) */
    235                + MULTIPLY( z2, FIX_1_451774981 )/* sqrt(2) * (c3+c7) */
    236                + MULTIPLY( z3, -FIX_2_172734803 )/* sqrt(2) * (-c1-c5) */
    237                + MULTIPLY( z4, FIX_1_061594337 );/* sqrt(2) * (c5+c7) */
    238 
    239         tmp2 = MULTIPLY( z1, -FIX_0_509795579 )/* sqrt(2) * (c7-c5) */
    240                + MULTIPLY( z2, -FIX_0_601344887 )/* sqrt(2) * (c5-c1) */
    241                + MULTIPLY( z3, FIX_0_899976223 )/* sqrt(2) * (c3-c7) */
    242                + MULTIPLY( z4, FIX_2_562915447 );/* sqrt(2) * (c1+c3) */
    243 
    244         /* Final output stage */
    245 
    246         outptr[0] = range_limit[(int) DESCALE( tmp10 + tmp2,
    247                                                CONST_BITS + PASS1_BITS + 3 + 1 )
    248                                 & RANGE_MASK];
    249         outptr[3] = range_limit[(int) DESCALE( tmp10 - tmp2,
    250                                                CONST_BITS + PASS1_BITS + 3 + 1 )
    251                                 & RANGE_MASK];
    252         outptr[1] = range_limit[(int) DESCALE( tmp12 + tmp0,
    253                                                CONST_BITS + PASS1_BITS + 3 + 1 )
    254                                 & RANGE_MASK];
    255         outptr[2] = range_limit[(int) DESCALE( tmp12 - tmp0,
    256                                                CONST_BITS + PASS1_BITS + 3 + 1 )
    257                                 & RANGE_MASK];
    258 
    259         wsptr += DCTSIZE;   /* advance pointer to next row */
    260     }
    261 }
    262 
    263 
    264 /*
    265  * Perform dequantization and inverse DCT on one block of coefficients,
    266  * producing a reduced-size 2x2 output block.
    267  */
    268 
    269 GLOBAL void
    270 jpeg_idct_2x2( j_decompress_ptr cinfo, jpeg_component_info * compptr,
    271                JCOEFPTR coef_block,
    272                JSAMPARRAY output_buf, JDIMENSION output_col ) {
    273     INT32 tmp0, tmp10, z1;
    274     JCOEFPTR inptr;
    275     ISLOW_MULT_TYPE * quantptr;
    276     int * wsptr;
    277     JSAMPROW outptr;
    278     JSAMPLE * range_limit = IDCT_range_limit( cinfo );
    279     int ctr;
    280     int workspace[DCTSIZE * 2];/* buffers data between passes */
    281     SHIFT_TEMPS
    282 
    283     /* Pass 1: process columns from input, store into work array. */
    284 
    285     inptr = coef_block;
    286     quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
    287     wsptr = workspace;
    288     for ( ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr-- ) {
    289         /* Don't bother to process columns 2,4,6 */
    290         if ( ( ctr == DCTSIZE - 2 ) || ( ctr == DCTSIZE - 4 ) || ( ctr == DCTSIZE - 6 ) ) {
    291             continue;
    292         }
    293         if ( ( inptr[DCTSIZE * 1] | inptr[DCTSIZE * 3] |
    294                inptr[DCTSIZE * 5] | inptr[DCTSIZE * 7] ) == 0 ) {
    295             /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
    296             int dcval = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] ) << PASS1_BITS;
    297 
    298             wsptr[DCTSIZE * 0] = dcval;
    299             wsptr[DCTSIZE * 1] = dcval;
    300 
    301             continue;
    302         }
    303 
    304         /* Even part */
    305 
    306         z1 = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] );
    307         tmp10 = z1 << ( CONST_BITS + 2 );
    308 
    309         /* Odd part */
    310 
    311         z1 = DEQUANTIZE( inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7] );
    312         tmp0 = MULTIPLY( z1, -FIX_0_720959822 );/* sqrt(2) * (c7-c5+c3-c1) */
    313         z1 = DEQUANTIZE( inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5] );
    314         tmp0 += MULTIPLY( z1, FIX_0_850430095 );/* sqrt(2) * (-c1+c3+c5+c7) */
    315         z1 = DEQUANTIZE( inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3] );
    316         tmp0 += MULTIPLY( z1, -FIX_1_272758580 );/* sqrt(2) * (-c1+c3-c5-c7) */
    317         z1 = DEQUANTIZE( inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1] );
    318         tmp0 += MULTIPLY( z1, FIX_3_624509785 );/* sqrt(2) * (c1+c3+c5+c7) */
    319 
    320         /* Final output stage */
    321 
    322         wsptr[DCTSIZE * 0] = (int) DESCALE( tmp10 + tmp0, CONST_BITS - PASS1_BITS + 2 );
    323         wsptr[DCTSIZE * 1] = (int) DESCALE( tmp10 - tmp0, CONST_BITS - PASS1_BITS + 2 );
    324     }
    325 
    326     /* Pass 2: process 2 rows from work array, store into output array. */
    327 
    328     wsptr = workspace;
    329     for ( ctr = 0; ctr < 2; ctr++ ) {
    330         outptr = output_buf[ctr] + output_col;
    331         /* It's not clear whether a zero row test is worthwhile here ... */
    332 
    333 #ifndef NO_ZERO_ROW_TEST
    334         if ( ( wsptr[1] | wsptr[3] | wsptr[5] | wsptr[7] ) == 0 ) {
    335             /* AC terms all zero */
    336             JSAMPLE dcval = range_limit[(int) DESCALE( (INT32) wsptr[0], PASS1_BITS + 3 )
    337                                         & RANGE_MASK];
    338 
    339             outptr[0] = dcval;
    340             outptr[1] = dcval;
    341 
    342             wsptr += DCTSIZE;/* advance pointer to next row */
    343             continue;
    344         }
    345 #endif
    346 
    347         /* Even part */
    348 
    349         tmp10 = ( (INT32) wsptr[0] ) << ( CONST_BITS + 2 );
    350 
    351         /* Odd part */
    352 
    353         tmp0 = MULTIPLY( (INT32) wsptr[7], -FIX_0_720959822 )/* sqrt(2) * (c7-c5+c3-c1) */
    354                + MULTIPLY( (INT32) wsptr[5], FIX_0_850430095 )/* sqrt(2) * (-c1+c3+c5+c7) */
    355                + MULTIPLY( (INT32) wsptr[3], -FIX_1_272758580 )/* sqrt(2) * (-c1+c3-c5-c7) */
    356                + MULTIPLY( (INT32) wsptr[1], FIX_3_624509785 );/* sqrt(2) * (c1+c3+c5+c7) */
    357 
    358         /* Final output stage */
    359 
    360         outptr[0] = range_limit[(int) DESCALE( tmp10 + tmp0,
    361                                                CONST_BITS + PASS1_BITS + 3 + 2 )
    362                                 & RANGE_MASK];
    363         outptr[1] = range_limit[(int) DESCALE( tmp10 - tmp0,
    364                                                CONST_BITS + PASS1_BITS + 3 + 2 )
    365                                 & RANGE_MASK];
    366 
    367         wsptr += DCTSIZE;   /* advance pointer to next row */
    368     }
    369 }
    370 
    371 
    372 /*
    373  * Perform dequantization and inverse DCT on one block of coefficients,
    374  * producing a reduced-size 1x1 output block.
    375  */
    376 
    377 GLOBAL void
    378 jpeg_idct_1x1( j_decompress_ptr cinfo, jpeg_component_info * compptr,
    379                JCOEFPTR coef_block,
    380                JSAMPARRAY output_buf, JDIMENSION output_col ) {
    381     int dcval;
    382     ISLOW_MULT_TYPE * quantptr;
    383     JSAMPLE * range_limit = IDCT_range_limit( cinfo );
    384     SHIFT_TEMPS
    385 
    386     /* We hardly need an inverse DCT routine for this: just take the
    387      * average pixel value, which is one-eighth of the DC coefficient.
    388      */
    389     quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
    390     dcval = DEQUANTIZE( coef_block[0], quantptr[0] );
    391     dcval = (int) DESCALE( (INT32) dcval, 3 );
    392 
    393     output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
    394 }
    395 
    396 #endif /* IDCT_SCALING_SUPPORTED */