DOOM-3-BFG

DOOM 3 BFG Edition
Log | Files | Refs

jidctint.cpp (16274B)


      1 /*
      2  * jidctint.c
      3  *
      4  * Copyright (C) 1991-1994, 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 a slow-but-accurate integer implementation of the
      9  * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
     10  * must also perform dequantization of the input coefficients.
     11  *
     12  * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
     13  * on each row (or vice versa, but it's more convenient to emit a row at
     14  * a time).  Direct algorithms are also available, but they are much more
     15  * complex and seem not to be any faster when reduced to code.
     16  *
     17  * This implementation is based on an algorithm described in
     18  *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
     19  *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
     20  *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
     21  * The primary algorithm described there uses 11 multiplies and 29 adds.
     22  * We use their alternate method with 12 multiplies and 32 adds.
     23  * The advantage of this method is that no data path contains more than one
     24  * multiplication; this allows a very simple and accurate implementation in
     25  * scaled fixed-point arithmetic, with a minimal number of shifts.
     26  */
     27 
     28 #define JPEG_INTERNALS
     29 #include "jinclude.h"
     30 #include "jpeglib.h"
     31 #include "jdct.h"        /* Private declarations for DCT subsystem */
     32 
     33 #ifdef DCT_ISLOW_SUPPORTED
     34 
     35 
     36 /*
     37  * This module is specialized to the case DCTSIZE = 8.
     38  */
     39 
     40 #if DCTSIZE != 8
     41 Sorry, this code only copes with 8 x8 DCTs.  /* deliberate syntax err */
     42     #endif
     43 
     44 
     45 /*
     46  * The poop on this scaling stuff is as follows:
     47  *
     48  * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
     49  * larger than the true IDCT outputs.  The final outputs are therefore
     50  * a factor of N larger than desired; since N=8 this can be cured by
     51  * a simple right shift at the end of the algorithm.  The advantage of
     52  * this arrangement is that we save two multiplications per 1-D IDCT,
     53  * because the y0 and y4 inputs need not be divided by sqrt(N).
     54  *
     55  * We have to do addition and subtraction of the integer inputs, which
     56  * is no problem, and multiplication by fractional constants, which is
     57  * a problem to do in integer arithmetic.  We multiply all the constants
     58  * by CONST_SCALE and convert them to integer constants (thus retaining
     59  * CONST_BITS bits of precision in the constants).  After doing a
     60  * multiplication we have to divide the product by CONST_SCALE, with proper
     61  * rounding, to produce the correct output.  This division can be done
     62  * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
     63  * as long as possible so that partial sums can be added together with
     64  * full fractional precision.
     65  *
     66  * The outputs of the first pass are scaled up by PASS1_BITS bits so that
     67  * they are represented to better-than-integral precision.  These outputs
     68  * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
     69  * with the recommended scaling.  (To scale up 12-bit sample data further, an
     70  * intermediate INT32 array would be needed.)
     71  *
     72  * To avoid overflow of the 32-bit intermediate results in pass 2, we must
     73  * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
     74  * shows that the values given below are the most effective.
     75  */
     76 
     77 #if BITS_IN_JSAMPLE == 8
     78 #define CONST_BITS  13
     79 #define PASS1_BITS  2
     80 #else
     81 #define CONST_BITS  13
     82 #define PASS1_BITS  1       /* lose a little precision to avoid overflow */
     83 #endif
     84 
     85 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
     86  * causing a lot of useless floating-point operations at run time.
     87  * To get around this we use the following pre-calculated constants.
     88  * If you change CONST_BITS you may want to add appropriate values.
     89  * (With a reasonable C compiler, you can just rely on the FIX() macro...)
     90  */
     91 
     92 #if CONST_BITS == 13
     93 #define FIX_0_298631336  ( (INT32)  2446 )    /* FIX(0.298631336) */
     94 #define FIX_0_390180644  ( (INT32)  3196 )    /* FIX(0.390180644) */
     95 #define FIX_0_541196100  ( (INT32)  4433 )    /* FIX(0.541196100) */
     96 #define FIX_0_765366865  ( (INT32)  6270 )    /* FIX(0.765366865) */
     97 #define FIX_0_899976223  ( (INT32)  7373 )    /* FIX(0.899976223) */
     98 #define FIX_1_175875602  ( (INT32)  9633 )    /* FIX(1.175875602) */
     99 #define FIX_1_501321110  ( (INT32)  12299 )   /* FIX(1.501321110) */
    100 #define FIX_1_847759065  ( (INT32)  15137 )   /* FIX(1.847759065) */
    101 #define FIX_1_961570560  ( (INT32)  16069 )   /* FIX(1.961570560) */
    102 #define FIX_2_053119869  ( (INT32)  16819 )   /* FIX(2.053119869) */
    103 #define FIX_2_562915447  ( (INT32)  20995 )   /* FIX(2.562915447) */
    104 #define FIX_3_072711026  ( (INT32)  25172 )   /* FIX(3.072711026) */
    105 #else
    106 #define FIX_0_298631336  FIX( 0.298631336 )
    107 #define FIX_0_390180644  FIX( 0.390180644 )
    108 #define FIX_0_541196100  FIX( 0.541196100 )
    109 #define FIX_0_765366865  FIX( 0.765366865 )
    110 #define FIX_0_899976223  FIX( 0.899976223 )
    111 #define FIX_1_175875602  FIX( 1.175875602 )
    112 #define FIX_1_501321110  FIX( 1.501321110 )
    113 #define FIX_1_847759065  FIX( 1.847759065 )
    114 #define FIX_1_961570560  FIX( 1.961570560 )
    115 #define FIX_2_053119869  FIX( 2.053119869 )
    116 #define FIX_2_562915447  FIX( 2.562915447 )
    117 #define FIX_3_072711026  FIX( 3.072711026 )
    118 #endif
    119 
    120 
    121 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
    122  * For 8-bit samples with the recommended scaling, all the variable
    123  * and constant values involved are no more than 16 bits wide, so a
    124  * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
    125  * For 12-bit samples, a full 32-bit multiplication will be needed.
    126  */
    127 
    128 #if BITS_IN_JSAMPLE == 8
    129 #define MULTIPLY( var, const )  MULTIPLY16C16( var, const )
    130 #else
    131 #define MULTIPLY( var, const )  ( ( var ) * ( const ) )
    132 #endif
    133 
    134 
    135 /* Dequantize a coefficient by multiplying it by the multiplier-table
    136  * entry; produce an int result.  In this module, both inputs and result
    137  * are 16 bits or less, so either int or short multiply will work.
    138  */
    139 
    140 #define DEQUANTIZE( coef, quantval )  ( ( (ISLOW_MULT_TYPE) ( coef ) ) * ( quantval ) )
    141 
    142 
    143 /*
    144  * Perform dequantization and inverse DCT on one block of coefficients.
    145  */
    146 
    147 GLOBAL void
    148 jpeg_idct_islow( j_decompress_ptr cinfo, jpeg_component_info * compptr,
    149                  JCOEFPTR coef_block,
    150                  JSAMPARRAY output_buf, JDIMENSION output_col ) {
    151     INT32 tmp0, tmp1, tmp2, tmp3;
    152     INT32 tmp10, tmp11, tmp12, tmp13;
    153     INT32 z1, z2, z3, z4, z5;
    154     JCOEFPTR inptr;
    155     ISLOW_MULT_TYPE * quantptr;
    156     int * wsptr;
    157     JSAMPROW outptr;
    158     JSAMPLE * range_limit = IDCT_range_limit( cinfo );
    159     int ctr;
    160     int workspace[DCTSIZE2];/* buffers data between passes */
    161     SHIFT_TEMPS
    162 
    163     /* Pass 1: process columns from input, store into work array. */
    164     /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
    165     /* furthermore, we scale the results by 2**PASS1_BITS. */
    166 
    167     inptr = coef_block;
    168     quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
    169     wsptr = workspace;
    170     for ( ctr = DCTSIZE; ctr > 0; ctr-- ) {
    171         /* Due to quantization, we will usually find that many of the input
    172          * coefficients are zero, especially the AC terms.  We can exploit this
    173          * by short-circuiting the IDCT calculation for any column in which all
    174          * the AC terms are zero.  In that case each output is equal to the
    175          * DC coefficient (with scale factor as needed).
    176          * With typical images and quantization tables, half or more of the
    177          * column DCT calculations can be simplified this way.
    178          */
    179 
    180         if ( ( inptr[DCTSIZE * 1] | inptr[DCTSIZE * 2] | inptr[DCTSIZE * 3] |
    181                inptr[DCTSIZE * 4] | inptr[DCTSIZE * 5] | inptr[DCTSIZE * 6] |
    182                inptr[DCTSIZE * 7] ) == 0 ) {
    183             /* AC terms all zero */
    184             int dcval = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] ) << PASS1_BITS;
    185 
    186             wsptr[DCTSIZE * 0] = dcval;
    187             wsptr[DCTSIZE * 1] = dcval;
    188             wsptr[DCTSIZE * 2] = dcval;
    189             wsptr[DCTSIZE * 3] = dcval;
    190             wsptr[DCTSIZE * 4] = dcval;
    191             wsptr[DCTSIZE * 5] = dcval;
    192             wsptr[DCTSIZE * 6] = dcval;
    193             wsptr[DCTSIZE * 7] = dcval;
    194 
    195             inptr++;    /* advance pointers to next column */
    196             quantptr++;
    197             wsptr++;
    198             continue;
    199         }
    200 
    201         /* Even part: reverse the even part of the forward DCT. */
    202         /* The rotator is sqrt(2)*c(-6). */
    203 
    204         z2 = DEQUANTIZE( inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2] );
    205         z3 = DEQUANTIZE( inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6] );
    206 
    207         z1 = MULTIPLY( z2 + z3, FIX_0_541196100 );
    208         tmp2 = z1 + MULTIPLY( z3, -FIX_1_847759065 );
    209         tmp3 = z1 + MULTIPLY( z2, FIX_0_765366865 );
    210 
    211         z2 = DEQUANTIZE( inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] );
    212         z3 = DEQUANTIZE( inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4] );
    213 
    214         tmp0 = ( z2 + z3 ) << CONST_BITS;
    215         tmp1 = ( z2 - z3 ) << CONST_BITS;
    216 
    217         tmp10 = tmp0 + tmp3;
    218         tmp13 = tmp0 - tmp3;
    219         tmp11 = tmp1 + tmp2;
    220         tmp12 = tmp1 - tmp2;
    221 
    222         /* Odd part per figure 8; the matrix is unitary and hence its
    223          * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
    224          */
    225 
    226         tmp0 = DEQUANTIZE( inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7] );
    227         tmp1 = DEQUANTIZE( inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5] );
    228         tmp2 = DEQUANTIZE( inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3] );
    229         tmp3 = DEQUANTIZE( inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1] );
    230 
    231         z1 = tmp0 + tmp3;
    232         z2 = tmp1 + tmp2;
    233         z3 = tmp0 + tmp2;
    234         z4 = tmp1 + tmp3;
    235         z5 = MULTIPLY( z3 + z4, FIX_1_175875602 );/* sqrt(2) * c3 */
    236 
    237         tmp0 = MULTIPLY( tmp0, FIX_0_298631336 );/* sqrt(2) * (-c1+c3+c5-c7) */
    238         tmp1 = MULTIPLY( tmp1, FIX_2_053119869 );/* sqrt(2) * ( c1+c3-c5+c7) */
    239         tmp2 = MULTIPLY( tmp2, FIX_3_072711026 );/* sqrt(2) * ( c1+c3+c5-c7) */
    240         tmp3 = MULTIPLY( tmp3, FIX_1_501321110 );/* sqrt(2) * ( c1+c3-c5-c7) */
    241         z1 = MULTIPLY( z1, -FIX_0_899976223 );/* sqrt(2) * (c7-c3) */
    242         z2 = MULTIPLY( z2, -FIX_2_562915447 );/* sqrt(2) * (-c1-c3) */
    243         z3 = MULTIPLY( z3, -FIX_1_961570560 );/* sqrt(2) * (-c3-c5) */
    244         z4 = MULTIPLY( z4, -FIX_0_390180644 );/* sqrt(2) * (c5-c3) */
    245 
    246         z3 += z5;
    247         z4 += z5;
    248 
    249         tmp0 += z1 + z3;
    250         tmp1 += z2 + z4;
    251         tmp2 += z2 + z3;
    252         tmp3 += z1 + z4;
    253 
    254         /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
    255 
    256         wsptr[DCTSIZE * 0] = (int) DESCALE( tmp10 + tmp3, CONST_BITS - PASS1_BITS );
    257         wsptr[DCTSIZE * 7] = (int) DESCALE( tmp10 - tmp3, CONST_BITS - PASS1_BITS );
    258         wsptr[DCTSIZE * 1] = (int) DESCALE( tmp11 + tmp2, CONST_BITS - PASS1_BITS );
    259         wsptr[DCTSIZE * 6] = (int) DESCALE( tmp11 - tmp2, CONST_BITS - PASS1_BITS );
    260         wsptr[DCTSIZE * 2] = (int) DESCALE( tmp12 + tmp1, CONST_BITS - PASS1_BITS );
    261         wsptr[DCTSIZE * 5] = (int) DESCALE( tmp12 - tmp1, CONST_BITS - PASS1_BITS );
    262         wsptr[DCTSIZE * 3] = (int) DESCALE( tmp13 + tmp0, CONST_BITS - PASS1_BITS );
    263         wsptr[DCTSIZE * 4] = (int) DESCALE( tmp13 - tmp0, CONST_BITS - PASS1_BITS );
    264 
    265         inptr++;        /* advance pointers to next column */
    266         quantptr++;
    267         wsptr++;
    268     }
    269 
    270     /* Pass 2: process rows from work array, store into output array. */
    271     /* Note that we must descale the results by a factor of 8 == 2**3, */
    272     /* and also undo the PASS1_BITS scaling. */
    273 
    274     wsptr = workspace;
    275     for ( ctr = 0; ctr < DCTSIZE; ctr++ ) {
    276         outptr = output_buf[ctr] + output_col;
    277         /* Rows of zeroes can be exploited in the same way as we did with columns.
    278          * However, the column calculation has created many nonzero AC terms, so
    279          * the simplification applies less often (typically 5% to 10% of the time).
    280          * On machines with very fast multiplication, it's possible that the
    281          * test takes more time than it's worth.  In that case this section
    282          * may be commented out.
    283          */
    284 
    285 #ifndef NO_ZERO_ROW_TEST
    286         if ( ( wsptr[1] | wsptr[2] | wsptr[3] | wsptr[4] | wsptr[5] | wsptr[6] |
    287                wsptr[7] ) == 0 ) {
    288             /* AC terms all zero */
    289             JSAMPLE dcval = range_limit[(int) DESCALE( (INT32) wsptr[0], PASS1_BITS + 3 )
    290                                         & RANGE_MASK];
    291 
    292             outptr[0] = dcval;
    293             outptr[1] = dcval;
    294             outptr[2] = dcval;
    295             outptr[3] = dcval;
    296             outptr[4] = dcval;
    297             outptr[5] = dcval;
    298             outptr[6] = dcval;
    299             outptr[7] = dcval;
    300 
    301             wsptr += DCTSIZE;/* advance pointer to next row */
    302             continue;
    303         }
    304 #endif
    305 
    306         /* Even part: reverse the even part of the forward DCT. */
    307         /* The rotator is sqrt(2)*c(-6). */
    308 
    309         z2 = (INT32) wsptr[2];
    310         z3 = (INT32) wsptr[6];
    311 
    312         z1 = MULTIPLY( z2 + z3, FIX_0_541196100 );
    313         tmp2 = z1 + MULTIPLY( z3, -FIX_1_847759065 );
    314         tmp3 = z1 + MULTIPLY( z2, FIX_0_765366865 );
    315 
    316         tmp0 = ( (INT32) wsptr[0] + (INT32) wsptr[4] ) << CONST_BITS;
    317         tmp1 = ( (INT32) wsptr[0] - (INT32) wsptr[4] ) << CONST_BITS;
    318 
    319         tmp10 = tmp0 + tmp3;
    320         tmp13 = tmp0 - tmp3;
    321         tmp11 = tmp1 + tmp2;
    322         tmp12 = tmp1 - tmp2;
    323 
    324         /* Odd part per figure 8; the matrix is unitary and hence its
    325          * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
    326          */
    327 
    328         tmp0 = (INT32) wsptr[7];
    329         tmp1 = (INT32) wsptr[5];
    330         tmp2 = (INT32) wsptr[3];
    331         tmp3 = (INT32) wsptr[1];
    332 
    333         z1 = tmp0 + tmp3;
    334         z2 = tmp1 + tmp2;
    335         z3 = tmp0 + tmp2;
    336         z4 = tmp1 + tmp3;
    337         z5 = MULTIPLY( z3 + z4, FIX_1_175875602 );/* sqrt(2) * c3 */
    338 
    339         tmp0 = MULTIPLY( tmp0, FIX_0_298631336 );/* sqrt(2) * (-c1+c3+c5-c7) */
    340         tmp1 = MULTIPLY( tmp1, FIX_2_053119869 );/* sqrt(2) * ( c1+c3-c5+c7) */
    341         tmp2 = MULTIPLY( tmp2, FIX_3_072711026 );/* sqrt(2) * ( c1+c3+c5-c7) */
    342         tmp3 = MULTIPLY( tmp3, FIX_1_501321110 );/* sqrt(2) * ( c1+c3-c5-c7) */
    343         z1 = MULTIPLY( z1, -FIX_0_899976223 );/* sqrt(2) * (c7-c3) */
    344         z2 = MULTIPLY( z2, -FIX_2_562915447 );/* sqrt(2) * (-c1-c3) */
    345         z3 = MULTIPLY( z3, -FIX_1_961570560 );/* sqrt(2) * (-c3-c5) */
    346         z4 = MULTIPLY( z4, -FIX_0_390180644 );/* sqrt(2) * (c5-c3) */
    347 
    348         z3 += z5;
    349         z4 += z5;
    350 
    351         tmp0 += z1 + z3;
    352         tmp1 += z2 + z4;
    353         tmp2 += z2 + z3;
    354         tmp3 += z1 + z4;
    355 
    356         /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
    357 
    358         outptr[0] = range_limit[(int) DESCALE( tmp10 + tmp3,
    359                                                CONST_BITS + PASS1_BITS + 3 )
    360                                 & RANGE_MASK];
    361         outptr[7] = range_limit[(int) DESCALE( tmp10 - tmp3,
    362                                                CONST_BITS + PASS1_BITS + 3 )
    363                                 & RANGE_MASK];
    364         outptr[1] = range_limit[(int) DESCALE( tmp11 + tmp2,
    365                                                CONST_BITS + PASS1_BITS + 3 )
    366                                 & RANGE_MASK];
    367         outptr[6] = range_limit[(int) DESCALE( tmp11 - tmp2,
    368                                                CONST_BITS + PASS1_BITS + 3 )
    369                                 & RANGE_MASK];
    370         outptr[2] = range_limit[(int) DESCALE( tmp12 + tmp1,
    371                                                CONST_BITS + PASS1_BITS + 3 )
    372                                 & RANGE_MASK];
    373         outptr[5] = range_limit[(int) DESCALE( tmp12 - tmp1,
    374                                                CONST_BITS + PASS1_BITS + 3 )
    375                                 & RANGE_MASK];
    376         outptr[3] = range_limit[(int) DESCALE( tmp13 + tmp0,
    377                                                CONST_BITS + PASS1_BITS + 3 )
    378                                 & RANGE_MASK];
    379         outptr[4] = range_limit[(int) DESCALE( tmp13 - tmp0,
    380                                                CONST_BITS + PASS1_BITS + 3 )
    381                                 & RANGE_MASK];
    382 
    383         wsptr += DCTSIZE;   /* advance pointer to next row */
    384     }
    385 }
    386 
    387 #endif /* DCT_ISLOW_SUPPORTED */