1  /********************************************************************** <BR>


2  This file is part of Crack dot Com's free source code release of


3  Golgotha. <a href="http://www.crack.com/golgotha_release"> <BR> for


4  information about compiling & licensing issues visit this URL</a>


5  <PRE> If that doesn't help, contact Jonathan Clark at


6  golgotha_source@usa.net (Subject should have "GOLG" in it)


7  ***********************************************************************/


8 


9  /*


10  * jidctred.c


11  *


12  * Copyright (C) 19941996, Thomas G. Lane.


13  * This file is part of the Independent JPEG Group's software.


14  * For conditions of distribution and use, see the accompanying README file.


15  *


16  * This file contains inverseDCT routines that produce reducedsize output:


17  * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.


18  *


19  * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M)


20  * algorithm used in jidctint.c. We simply replace each 8to8 1D IDCT step


21  * with an 8to4 step that produces the four averages of two adjacent outputs


22  * (or an 8to2 step producing two averages of four outputs, for 2x2 output).


23  * These steps were derived by computing the corresponding values at the end


24  * of the normal LL&M code, then simplifying as much as possible.


25  *


26  * 1x1 is trivial: just take the DC coefficient divided by 8.


27  *


28  * See jidctint.c for additional comments.


29  */


30 


31  #define JPEG_INTERNALS


32  #include "loaders/jpg/jinclude.h"


33  #include "loaders/jpg/jpeglib.h"


34  #include "loaders/jpg/jdct.h" /* Private declarations for DCT subsystem */


35 


36  #ifdef IDCT_SCALING_SUPPORTED


37 


38 


39  /*


40  * This module is specialized to the case DCTSIZE = 8.


41  */


42 


43  #if DCTSIZE != 8


44  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */


45  #endif


46 


47 


48  /* Scaling is the same as in jidctint.c. */


49 


50  #if BITS_IN_JSAMPLE == 8


51  #define CONST_BITS 13


52  #define PASS1_BITS 2


53  #else


54  #define CONST_BITS 13


55  #define PASS1_BITS 1 /* lose a little precision to avoid overflow */


56  #endif


57 


58  /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus


59  * causing a lot of useless floatingpoint operations at run time.


60  * To get around this we use the following precalculated constants.


61  * If you change CONST_BITS you may want to add appropriate values.


62  * (With a reasonable C compiler, you can just rely on the FIX() macro...)


63  */


64 


65  #if CONST_BITS == 13


66  #define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */


67  #define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */


68  #define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */


69  #define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */


70  #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */


71  #define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */


72  #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */


73  #define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */


74  #define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */


75  #define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */


76  #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */


77  #define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */


78  #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */


79  #define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */


80  #else


81  #define FIX_0_211164243 FIX(0.211164243)


82  #define FIX_0_509795579 FIX(0.509795579)


83  #define FIX_0_601344887 FIX(0.601344887)


84  #define FIX_0_720959822 FIX(0.720959822)


85  #define FIX_0_765366865 FIX(0.765366865)


86  #define FIX_0_850430095 FIX(0.850430095)


87  #define FIX_0_899976223 FIX(0.899976223)


88  #define FIX_1_061594337 FIX(1.061594337)


89  #define FIX_1_272758580 FIX(1.272758580)


90  #define FIX_1_451774981 FIX(1.451774981)


91  #define FIX_1_847759065 FIX(1.847759065)


92  #define FIX_2_172734803 FIX(2.172734803)


93  #define FIX_2_562915447 FIX(2.562915447)


94  #define FIX_3_624509785 FIX(3.624509785)


95  #endif


96 


97 


98  /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.


99  * For 8bit samples with the recommended scaling, all the variable


100  * and constant values involved are no more than 16 bits wide, so a


101  * 16x16>32 bit multiply can be used instead of a full 32x32 multiply.


102  * For 12bit samples, a full 32bit multiplication will be needed.


103  */


104 


105  #if BITS_IN_JSAMPLE == 8


106  #define MULTIPLY(var,const) MULTIPLY16C16(var,const)


107  #else


108  #define MULTIPLY(var,const) ((var) * (const))


109  #endif


110 


111 


112  /* Dequantize a coefficient by multiplying it by the multipliertable


113  * entry; produce an int result. In this module, both inputs and result


114  * are 16 bits or less, so either int or short multiply will work.


115  */


116 


117  #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))


118 


119 


120  /*


121  * Perform dequantization and inverse DCT on one block of coefficients,


122  * producing a reducedsize 4x4 output block.


123  */


124 


125  GLOBAL(void)


126  jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,


127  JCOEFPTR coef_block,


128  JSAMPARRAY output_buf, JDIMENSION output_col)


129  {


130  INT32 tmp0, tmp2, tmp10, tmp12;


131  INT32 z1, z2, z3, z4;


132  JCOEFPTR inptr;


133  ISLOW_MULT_TYPE * quantptr;


134  int * wsptr;


135  JSAMPROW outptr;


136  JSAMPLE *range_limit = IDCT_range_limit(cinfo);


137  int ctr;


138  int workspace[DCTSIZE*4]; /* buffers data between passes */


139  SHIFT_TEMPS


140 


141  /* Pass 1: process columns from input, store into work array. */


142 


143  inptr = coef_block;


144  quantptr = (ISLOW_MULT_TYPE *) compptr>dct_table;


145  wsptr = workspace;


146  for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr) {


147  /* Don't bother to process column 4, because second pass won't use it */


148  if (ctr == DCTSIZE4)


149  continue;


150  if ((inptr[DCTSIZE*1]  inptr[DCTSIZE*2]  inptr[DCTSIZE*3] 


151  inptr[DCTSIZE*5]  inptr[DCTSIZE*6]  inptr[DCTSIZE*7]) == 0) {


152  /* AC terms all zero; we need not examine term 4 for 4x4 output */


153  int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;


154 


155  wsptr[DCTSIZE*0] = dcval;


156  wsptr[DCTSIZE*1] = dcval;


157  wsptr[DCTSIZE*2] = dcval;


158  wsptr[DCTSIZE*3] = dcval;


159 


160  continue;


161  }


162 


163  /* Even part */


164 


165  tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);


166  tmp0 <<= (CONST_BITS+1);


167 


168  z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);


169  z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);


170 


171  tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3,  FIX_0_765366865);


172 


173  tmp10 = tmp0 + tmp2;


174  tmp12 = tmp0  tmp2;


175 


176  /* Odd part */


177 


178  z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);


179  z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);


180  z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);


181  z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);


182 


183  tmp0 = MULTIPLY(z1,  FIX_0_211164243) /* sqrt(2) * (c3c1) */


184  + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */


185  + MULTIPLY(z3,  FIX_2_172734803) /* sqrt(2) * (c1c5) */


186  + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */


187 


188  tmp2 = MULTIPLY(z1,  FIX_0_509795579) /* sqrt(2) * (c7c5) */


189  + MULTIPLY(z2,  FIX_0_601344887) /* sqrt(2) * (c5c1) */


190  + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3c7) */


191  + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */


192 


193  /* Final output stage */


194 


195  wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITSPASS1_BITS+1);


196  wsptr[DCTSIZE*3] = (int) DESCALE(tmp10  tmp2, CONST_BITSPASS1_BITS+1);


197  wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITSPASS1_BITS+1);


198  wsptr[DCTSIZE*2] = (int) DESCALE(tmp12  tmp0, CONST_BITSPASS1_BITS+1);


199  }


200 


201  /* Pass 2: process 4 rows from work array, store into output array. */


202 


203  wsptr = workspace;


204  for (ctr = 0; ctr < 4; ctr++) {


205  outptr = output_buf[ctr] + output_col;


206  /* It's not clear whether a zero row test is worthwhile here ... */


207 


208  #ifndef NO_ZERO_ROW_TEST


209  if ((wsptr[1]  wsptr[2]  wsptr[3]  wsptr[5]  wsptr[6] 


210  wsptr[7]) == 0) {


211  /* AC terms all zero */


212  JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)


213  & RANGE_MASK];


214 


215  outptr[0] = dcval;


216  outptr[1] = dcval;


217  outptr[2] = dcval;


218  outptr[3] = dcval;


219 


220  wsptr += DCTSIZE; /* advance pointer to next row */


221  continue;


222  }


223  #endif


224 


225  /* Even part */


226 


227  tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);


228 


229  tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)


230  + MULTIPLY((INT32) wsptr[6],  FIX_0_765366865);


231 


232  tmp10 = tmp0 + tmp2;


233  tmp12 = tmp0  tmp2;


234 


235  /* Odd part */


236 


237  z1 = (INT32) wsptr[7];


238  z2 = (INT32) wsptr[5];


239  z3 = (INT32) wsptr[3];


240  z4 = (INT32) wsptr[1];


241 


242  tmp0 = MULTIPLY(z1,  FIX_0_211164243) /* sqrt(2) * (c3c1) */


243  + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */


244  + MULTIPLY(z3,  FIX_2_172734803) /* sqrt(2) * (c1c5) */


245  + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */


246 


247  tmp2 = MULTIPLY(z1,  FIX_0_509795579) /* sqrt(2) * (c7c5) */


248  + MULTIPLY(z2,  FIX_0_601344887) /* sqrt(2) * (c5c1) */


249  + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3c7) */


250  + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */


251 


252  /* Final output stage */


253 


254  outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,


255  CONST_BITS+PASS1_BITS+3+1)


256  & RANGE_MASK];


257  outptr[3] = range_limit[(int) DESCALE(tmp10  tmp2,


258  CONST_BITS+PASS1_BITS+3+1)


259  & RANGE_MASK];


260  outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,


261  CONST_BITS+PASS1_BITS+3+1)


262  & RANGE_MASK];


263  outptr[2] = range_limit[(int) DESCALE(tmp12  tmp0,


264  CONST_BITS+PASS1_BITS+3+1)


265  & RANGE_MASK];


266 


267  wsptr += DCTSIZE; /* advance pointer to next row */


268  }


269  }


270 


271 


272  /*


273  * Perform dequantization and inverse DCT on one block of coefficients,


274  * producing a reducedsize 2x2 output block.


275  */


276 


277  GLOBAL(void)


278  jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,


279  JCOEFPTR coef_block,


280  JSAMPARRAY output_buf, JDIMENSION output_col)


281  {


282  INT32 tmp0, tmp10, z1;


283  JCOEFPTR inptr;


284  ISLOW_MULT_TYPE * quantptr;


285  int * wsptr;


286  JSAMPROW outptr;


287  JSAMPLE *range_limit = IDCT_range_limit(cinfo);


288  int ctr;


289  int workspace[DCTSIZE*2]; /* buffers data between passes */


290  SHIFT_TEMPS


291 


292  /* Pass 1: process columns from input, store into work array. */


293 


294  inptr = coef_block;


295  quantptr = (ISLOW_MULT_TYPE *) compptr>dct_table;


296  wsptr = workspace;


297  for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr) {


298  /* Don't bother to process columns 2,4,6 */


299  if (ctr == DCTSIZE2  ctr == DCTSIZE4  ctr == DCTSIZE6)


300  continue;


301  if ((inptr[DCTSIZE*1]  inptr[DCTSIZE*3] 


302  inptr[DCTSIZE*5]  inptr[DCTSIZE*7]) == 0) {


303  /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */


304  int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;


305 


306  wsptr[DCTSIZE*0] = dcval;


307  wsptr[DCTSIZE*1] = dcval;


308 


309  continue;


310  }


311 


312  /* Even part */


313 


314  z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);


315  tmp10 = z1 << (CONST_BITS+2);


316 


317  /* Odd part */


318 


319  z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);


320  tmp0 = MULTIPLY(z1,  FIX_0_720959822); /* sqrt(2) * (c7c5+c3c1) */


321  z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);


322  tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (c1+c3+c5+c7) */


323  z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);


324  tmp0 += MULTIPLY(z1,  FIX_1_272758580); /* sqrt(2) * (c1+c3c5c7) */


325  z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);


326  tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */


327 


328  /* Final output stage */


329 


330  wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITSPASS1_BITS+2);


331  wsptr[DCTSIZE*1] = (int) DESCALE(tmp10  tmp0, CONST_BITSPASS1_BITS+2);


332  }


333 


334  /* Pass 2: process 2 rows from work array, store into output array. */


335 


336  wsptr = workspace;


337  for (ctr = 0; ctr < 2; ctr++) {


338  outptr = output_buf[ctr] + output_col;


339  /* It's not clear whether a zero row test is worthwhile here ... */


340 


341  #ifndef NO_ZERO_ROW_TEST


342  if ((wsptr[1]  wsptr[3]  wsptr[5]  wsptr[7]) == 0) {


343  /* AC terms all zero */


344  JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)


345  & RANGE_MASK];


346 


347  outptr[0] = dcval;


348  outptr[1] = dcval;


349 


350  wsptr += DCTSIZE; /* advance pointer to next row */


351  continue;


352  }


353  #endif


354 


355  /* Even part */


356 


357  tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);


358 


359  /* Odd part */


360 


361  tmp0 = MULTIPLY((INT32) wsptr[7],  FIX_0_720959822) /* sqrt(2) * (c7c5+c3c1) */


362  + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (c1+c3+c5+c7) */


363  + MULTIPLY((INT32) wsptr[3],  FIX_1_272758580) /* sqrt(2) * (c1+c3c5c7) */


364  + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */


365 


366  /* Final output stage */


367 


368  outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,


369  CONST_BITS+PASS1_BITS+3+2)


370  & RANGE_MASK];


371  outptr[1] = range_limit[(int) DESCALE(tmp10  tmp0,


372  CONST_BITS+PASS1_BITS+3+2)


373  & RANGE_MASK];


374 


375  wsptr += DCTSIZE; /* advance pointer to next row */


376  }


377  }


378 


379 


380  /*


381  * Perform dequantization and inverse DCT on one block of coefficients,


382  * producing a reducedsize 1x1 output block.


383  */


384 


385  GLOBAL(void)


386  jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,


387  JCOEFPTR coef_block,


388  JSAMPARRAY output_buf, JDIMENSION output_col)


389  {


390  int dcval;


391  ISLOW_MULT_TYPE * quantptr;


392  JSAMPLE *range_limit = IDCT_range_limit(cinfo);


393  SHIFT_TEMPS


394 


395  /* We hardly need an inverse DCT routine for this: just take the


396  * average pixel value, which is oneeighth of the DC coefficient.


397  */


398  quantptr = (ISLOW_MULT_TYPE *) compptr>dct_table;


399  dcval = DEQUANTIZE(coef_block[0], quantptr[0]);


400  dcval = (int) DESCALE((INT32) dcval, 3);


401 


402  output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];


403  }


404 


405  #endif /* IDCT_SCALING_SUPPORTED */

