src/jpeg/jcdctmgr.c

/*
 * jcdctmgr.c
 *
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
 * This file contains the forward-DCT management logic.
 * This code selects a particular DCT implementation to be used,
 * and it performs related housekeeping chores including coefficient
 * quantization.
 */

#define JPEG_INTERNALS
#include "jinclude.h"
#include "jpeglib.h"
#include "jdct.h"		/* Private declarations for DCT subsystem */


/* Private subobject for this module */

typedef struct {
  struct jpeg_forward_dct pub;	/* public fields */

  /* Pointer to the DCT routine actually in use */
  forward_DCT_method_ptr do_dct;

  /* The actual post-DCT divisors --- not identical to the quant table
   * entries, because of scaling (especially for an unnormalized DCT).
   * Each table is given in normal array order.
   */
  DCTELEM * divisors[NUM_QUANT_TBLS];

#ifdef DCT_FLOAT_SUPPORTED
  /* Same as above for the floating-point case. */
  float_DCT_method_ptr do_float_dct;
  FAST_FLOAT * float_divisors[NUM_QUANT_TBLS];
#endif
} my_fdct_controller;

typedef my_fdct_controller * my_fdct_ptr;


/*
 * Initialize for a processing pass.
 * Verify that all referenced Q-tables are present, and set up
 * the divisor table for each one.
 * In the current implementation, DCT of all components is done during
 * the first pass, even if only some components will be output in the
 * first scan.  Hence all components should be examined here.
 */

METHODDEF(void)
start_pass_fdctmgr (j_compress_ptr cinfo)
{
  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
  int ci, qtblno, i;
  jpeg_component_info *compptr;
  JQUANT_TBL * qtbl;
  DCTELEM * dtbl;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    qtblno = compptr->quant_tbl_no;
    /* Make sure specified quantization table is present */
    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
	cinfo->quant_tbl_ptrs[qtblno] == NULL)
      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
    qtbl = cinfo->quant_tbl_ptrs[qtblno];
    /* Compute divisors for this quant table */
    /* We may do this more than once for same table, but it's not a big deal */
    switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
    case JDCT_ISLOW:
      /* For LL&M IDCT method, divisors are equal to raw quantization
       * coefficients multiplied by 8 (to counteract scaling).
       */
      if (fdct->divisors[qtblno] == NULL) {
	fdct->divisors[qtblno] = (DCTELEM *)
	  (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
				      DCTSIZE2 * SIZEOF(DCTELEM));
      }
      dtbl = fdct->divisors[qtblno];
      for (i = 0; i < DCTSIZE2; i++) {
	dtbl[i] = ((DCTELEM) qtbl->quantval[i]) << 3;
      }
      break;
#endif
#ifdef DCT_IFAST_SUPPORTED
    case JDCT_IFAST:
      {
	/* For AA&N IDCT method, divisors are equal to quantization
	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
	 *   scalefactor[0] = 1
	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
	 * We apply a further scale factor of 8.
	 */
#define CONST_BITS 14
	static const INT16 aanscales[DCTSIZE2] = {
	  /* precomputed values scaled up by 14 bits */
	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
	};
	SHIFT_TEMPS

	if (fdct->divisors[qtblno] == NULL) {
	  fdct->divisors[qtblno] = (DCTELEM *)
	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
					DCTSIZE2 * SIZEOF(DCTELEM));
	}
	dtbl = fdct->divisors[qtblno];
	for (i = 0; i < DCTSIZE2; i++) {
	  dtbl[i] = (DCTELEM)
	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
				  (INT32) aanscales[i]),
		    CONST_BITS-3);
	}
      }
      break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
    case JDCT_FLOAT:
      {
	/* For float AA&N IDCT method, divisors are equal to quantization
	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
	 *   scalefactor[0] = 1
	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
	 * We apply a further scale factor of 8.
	 * What's actually stored is 1/divisor so that the inner loop can
	 * use a multiplication rather than a division.
	 */
	FAST_FLOAT * fdtbl;
	int row, col;
	static const double aanscalefactor[DCTSIZE] = {
	  1.0, 1.387039845, 1.306562965, 1.175875602,
	  1.0, 0.785694958, 0.541196100, 0.275899379
	};

	if (fdct->float_divisors[qtblno] == NULL) {
	  fdct->float_divisors[qtblno] = (FAST_FLOAT *)
	    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
					DCTSIZE2 * SIZEOF(FAST_FLOAT));
	}
	fdtbl = fdct->float_divisors[qtblno];
	i = 0;
	for (row = 0; row < DCTSIZE; row++) {
	  for (col = 0; col < DCTSIZE; col++) {
	    fdtbl[i] = (FAST_FLOAT)
	      (1.0 / (((double) qtbl->quantval[i] *
		       aanscalefactor[row] * aanscalefactor[col] * 8.0)));
	    i++;
	  }
	}
      }
      break;
#endif
    default:
      ERREXIT(cinfo, JERR_NOT_COMPILED);
      break;
    }
  }
}


/*
 * Perform forward DCT on one or more blocks of a component.
 *
 * The input samples are taken from the sample_data[] array starting at
 * position start_row/start_col, and moving to the right for any additional
 * blocks. The quantized coefficients are returned in coef_blocks[].
 */

METHODDEF(void)
forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
	     JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
	     JDIMENSION start_row, JDIMENSION start_col,
	     JDIMENSION num_blocks)
/* This version is used for integer DCT implementations. */
{
  /* This routine is heavily used, so it's worth coding it tightly. */
  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
  forward_DCT_method_ptr do_dct = fdct->do_dct;
  DCTELEM * divisors = fdct->divisors[compptr->quant_tbl_no];
  DCTELEM workspace[DCTSIZE2];	/* work area for FDCT subroutine */
  JDIMENSION bi;

  sample_data += start_row;	/* fold in the vertical offset once */

  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
    /* Load data into workspace, applying unsigned->signed conversion */
    { register DCTELEM *workspaceptr;
      register JSAMPROW elemptr;
      register int elemr;

      workspaceptr = workspace;
      for (elemr = 0; elemr < DCTSIZE; elemr++) {
	elemptr = sample_data[elemr] + start_col;
#if DCTSIZE == 8		/* unroll the inner loop */
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	*workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
#else
	{ register int elemc;
	  for (elemc = DCTSIZE; elemc > 0; elemc--) {
	    *workspaceptr++ = GETJSAMPLE(*elemptr++) - CENTERJSAMPLE;
	  }
	}
#endif
      }
    }

    /* Perform the DCT */
    (*do_dct) (workspace);

    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    { register DCTELEM temp, qval;
      register int i;
      register JCOEFPTR output_ptr = coef_blocks[bi];

      for (i = 0; i < DCTSIZE2; i++) {
	qval = divisors[i];
	temp = workspace[i];
	/* Divide the coefficient value by qval, ensuring proper rounding.
	 * Since C does not specify the direction of rounding for negative
	 * quotients, we have to force the dividend positive for portability.
	 *
	 * In most files, at least half of the output values will be zero
	 * (at default quantization settings, more like three-quarters...)
	 * so we should ensure that this case is fast.  On many machines,
	 * a comparison is enough cheaper than a divide to make a special test
	 * a win.  Since both inputs will be nonnegative, we need only test
	 * for a < b to discover whether a/b is 0.
	 * If your machine's division is fast enough, define FAST_DIVIDE.
	 */
#ifdef FAST_DIVIDE
#define DIVIDE_BY(a,b)	a /= b
#else
#define DIVIDE_BY(a,b)	if (a >= b) a /= b; else a = 0
#endif
	if (temp < 0) {
	  temp = -temp;
	  temp += qval>>1;	/* for rounding */
	  DIVIDE_BY(temp, qval);
	  temp = -temp;
	} else {
	  temp += qval>>1;	/* for rounding */
	  DIVIDE_BY(temp, qval);
	}
	output_ptr[i] = (JCOEF) temp;
      }
    }
  }
}


#ifdef DCT_FLOAT_SUPPORTED

METHODDEF(void)
forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
		   JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
		   JDIMENSION start_row, JDIMENSION start_col,
		   JDIMENSION num_blocks)
/* This version is used for floating-point DCT implementations. */
{
  /* This routine is heavily used, so it's worth coding it tightly. */
  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
  float_DCT_method_ptr do_dct = fdct->do_float_dct;
  FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
  FAST_FLOAT workspace[DCTSIZE2]; /* work area for FDCT subroutine */
  JDIMENSION bi;

  sample_data += start_row;	/* fold in the vertical offset once */

  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
    /* Load data into workspace, applying unsigned->signed conversion */
    { register FAST_FLOAT *workspaceptr;
      register JSAMPROW elemptr;
      register int elemr;

      workspaceptr = workspace;
      for (elemr = 0; elemr < DCTSIZE; elemr++) {
	elemptr = sample_data[elemr] + start_col;
#if DCTSIZE == 8		/* unroll the inner loop */
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	*workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
#else
	{ register int elemc;
	  for (elemc = DCTSIZE; elemc > 0; elemc--) {
	    *workspaceptr++ = (FAST_FLOAT)
	      (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	  }
	}
#endif
      }
    }

    /* Perform the DCT */
    (*do_dct) (workspace);

    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    { register FAST_FLOAT temp;
      register int i;
      register JCOEFPTR output_ptr = coef_blocks[bi];

      for (i = 0; i < DCTSIZE2; i++) {
	/* Apply the quantization and scaling factor */
	temp = workspace[i] * divisors[i];
	/* Round to nearest integer.
	 * Since C does not specify the direction of rounding for negative
	 * quotients, we have to force the dividend positive for portability.
	 * The maximum coefficient size is +-16K (for 12-bit data), so this
	 * code should work for either 16-bit or 32-bit ints.
	 */
	output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
      }
    }
  }
}

#endif /* DCT_FLOAT_SUPPORTED */


/*
 * Initialize FDCT manager.
 */

GLOBAL(void)
jinit_forward_dct (j_compress_ptr cinfo)
{
  my_fdct_ptr fdct;
  int i;

  fdct = (my_fdct_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
				SIZEOF(my_fdct_controller));
  cinfo->fdct = (struct jpeg_forward_dct *) fdct;
  fdct->pub.start_pass = start_pass_fdctmgr;

  switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
  case JDCT_ISLOW:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_islow;
    break;
#endif
#ifdef DCT_IFAST_SUPPORTED
  case JDCT_IFAST:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_ifast;
    break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
  case JDCT_FLOAT:
    fdct->pub.forward_DCT = forward_DCT_float;
    fdct->do_float_dct = jpeg_fdct_float;
    break;
#endif
  default:
    ERREXIT(cinfo, JERR_NOT_COMPILED);
    break;
  }

  /* Mark divisor tables unallocated */
  for (i = 0; i < NUM_QUANT_TBLS; i++) {
    fdct->divisors[i] = NULL;
#ifdef DCT_FLOAT_SUPPORTED
    fdct->float_divisors[i] = NULL;
#endif
  }
}
Commit	Line	Data
	1	/*
	2	* jcdctmgr.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 forward-DCT management logic.
	9	* This code selects a particular DCT implementation to be used,
	10	* and it performs related housekeeping chores including coefficient
	11	* quantization.
	12	*/
	13
	14	#define JPEG_INTERNALS
	15	#include "jinclude.h"
	16	#include "jpeglib.h"
	17	#include "jdct.h" /* Private declarations for DCT subsystem */
	18
	19
	20	/* Private subobject for this module */
	21
	22	typedef struct {
	23	struct jpeg_forward_dct pub; /* public fields */
	24
	25	/* Pointer to the DCT routine actually in use */
	26	forward_DCT_method_ptr do_dct;
	27
	28	/* The actual post-DCT divisors --- not identical to the quant table
	29	* entries, because of scaling (especially for an unnormalized DCT).
	30	* Each table is given in normal array order.
	31	*/
	32	DCTELEM * divisors[NUM_QUANT_TBLS];
	33
	34	#ifdef DCT_FLOAT_SUPPORTED
	35	/* Same as above for the floating-point case. */
	36	float_DCT_method_ptr do_float_dct;
	37	FAST_FLOAT * float_divisors[NUM_QUANT_TBLS];
	38	#endif
	39	} my_fdct_controller;
	40
	41	typedef my_fdct_controller * my_fdct_ptr;
	42
	43
	44	/*
	45	* Initialize for a processing pass.
	46	* Verify that all referenced Q-tables are present, and set up
	47	* the divisor table for each one.
	48	* In the current implementation, DCT of all components is done during
	49	* the first pass, even if only some components will be output in the
	50	* first scan. Hence all components should be examined here.
	51	*/
	52
	53	METHODDEF(void)
	54	start_pass_fdctmgr (j_compress_ptr cinfo)
	55	{
	56	my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
	57	int ci, qtblno, i;
	58	jpeg_component_info *compptr;
	59	JQUANT_TBL * qtbl;
	60	DCTELEM * dtbl;
	61
	62	for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
	63	ci++, compptr++) {
	64	qtblno = compptr->quant_tbl_no;
	65	/* Make sure specified quantization table is present */
	66	if (qtblno < 0 \|\| qtblno >= NUM_QUANT_TBLS \|\|
	67	cinfo->quant_tbl_ptrs[qtblno] == NULL)
	68	ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
	69	qtbl = cinfo->quant_tbl_ptrs[qtblno];
	70	/* Compute divisors for this quant table */
	71	/* We may do this more than once for same table, but it's not a big deal */
	72	switch (cinfo->dct_method) {
	73	#ifdef DCT_ISLOW_SUPPORTED
	74	case JDCT_ISLOW:
	75	/* For LL&M IDCT method, divisors are equal to raw quantization
	76	* coefficients multiplied by 8 (to counteract scaling).
	77	*/
	78	if (fdct->divisors[qtblno] == NULL) {
	79	fdct->divisors[qtblno] = (DCTELEM *)
	80	(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
	81	DCTSIZE2 * SIZEOF(DCTELEM));
	82	}
	83	dtbl = fdct->divisors[qtblno];
	84	for (i = 0; i < DCTSIZE2; i++) {
	85	dtbl[i] = ((DCTELEM) qtbl->quantval[i]) << 3;
	86	}
	87	break;
	88	#endif
	89	#ifdef DCT_IFAST_SUPPORTED
	90	case JDCT_IFAST:
	91	{
	92	/* For AA&N IDCT method, divisors are equal to quantization
	93	* coefficients scaled by scalefactor[row]*scalefactor[col], where
	94	* scalefactor[0] = 1
	95	* scalefactor[k] = cos(kPI/16) sqrt(2) for k=1..7
	96	* We apply a further scale factor of 8.
	97	*/
	98	#define CONST_BITS 14
	99	static const INT16 aanscales[DCTSIZE2] = {
	100	/* precomputed values scaled up by 14 bits */
	101	16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
	102	22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
	103	21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
	104	19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
	105	16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
	106	12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
	107	8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
	108	4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
	109	};
	110	SHIFT_TEMPS
	111
	112	if (fdct->divisors[qtblno] == NULL) {
	113	fdct->divisors[qtblno] = (DCTELEM *)
	114	(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
	115	DCTSIZE2 * SIZEOF(DCTELEM));
	116	}
	117	dtbl = fdct->divisors[qtblno];
	118	for (i = 0; i < DCTSIZE2; i++) {
	119	dtbl[i] = (DCTELEM)
	120	DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
	121	(INT32) aanscales[i]),
	122	CONST_BITS-3);
	123	}
	124	}
	125	break;
	126	#endif
	127	#ifdef DCT_FLOAT_SUPPORTED
	128	case JDCT_FLOAT:
	129	{
	130	/* For float AA&N IDCT method, divisors are equal to quantization
	131	* coefficients scaled by scalefactor[row]*scalefactor[col], where
	132	* scalefactor[0] = 1
	133	* scalefactor[k] = cos(kPI/16) sqrt(2) for k=1..7
	134	* We apply a further scale factor of 8.
	135	* What's actually stored is 1/divisor so that the inner loop can
	136	* use a multiplication rather than a division.
	137	*/
	138	FAST_FLOAT * fdtbl;
	139	int row, col;
	140	static const double aanscalefactor[DCTSIZE] = {
	141	1.0, 1.387039845, 1.306562965, 1.175875602,
	142	1.0, 0.785694958, 0.541196100, 0.275899379
	143	};
	144
	145	if (fdct->float_divisors[qtblno] == NULL) {
	146	fdct->float_divisors[qtblno] = (FAST_FLOAT *)
	147	(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
	148	DCTSIZE2 * SIZEOF(FAST_FLOAT));
	149	}
	150	fdtbl = fdct->float_divisors[qtblno];
	151	i = 0;
	152	for (row = 0; row < DCTSIZE; row++) {
	153	for (col = 0; col < DCTSIZE; col++) {
	154	fdtbl[i] = (FAST_FLOAT)
	155	(1.0 / (((double) qtbl->quantval[i] *
	156	aanscalefactor[row] * aanscalefactor[col] * 8.0)));
	157	i++;
	158	}
	159	}
	160	}
	161	break;
	162	#endif
	163	default:
	164	ERREXIT(cinfo, JERR_NOT_COMPILED);
	165	break;
	166	}
	167	}
	168	}
	169
	170
	171	/*
	172	* Perform forward DCT on one or more blocks of a component.
	173	*
	174	* The input samples are taken from the sample_data[] array starting at
	175	* position start_row/start_col, and moving to the right for any additional
	176	* blocks. The quantized coefficients are returned in coef_blocks[].
	177	*/
	178
	179	METHODDEF(void)
	180	forward_DCT (j_compress_ptr cinfo, jpeg_component_info * compptr,
	181	JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
	182	JDIMENSION start_row, JDIMENSION start_col,
	183	JDIMENSION num_blocks)
	184	/* This version is used for integer DCT implementations. */
	185	{
	186	/* This routine is heavily used, so it's worth coding it tightly. */
	187	my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
	188	forward_DCT_method_ptr do_dct = fdct->do_dct;
	189	DCTELEM * divisors = fdct->divisors[compptr->quant_tbl_no];
	190	DCTELEM workspace[DCTSIZE2]; /* work area for FDCT subroutine */
	191	JDIMENSION bi;
	192
	193	sample_data += start_row; /* fold in the vertical offset once */
	194
	195	for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
	196	/* Load data into workspace, applying unsigned->signed conversion */
	197	{ register DCTELEM *workspaceptr;
	198	register JSAMPROW elemptr;
	199	register int elemr;
	200
	201	workspaceptr = workspace;
	202	for (elemr = 0; elemr < DCTSIZE; elemr++) {
	203	elemptr = sample_data[elemr] + start_col;
	204	#if DCTSIZE == 8 /* unroll the inner loop */
	205	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	206	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	207	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	208	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	209	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	210	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	211	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	212	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	213	#else
	214	{ register int elemc;
	215	for (elemc = DCTSIZE; elemc > 0; elemc--) {
	216	workspaceptr++ = GETJSAMPLE(elemptr++) - CENTERJSAMPLE;
	217	}
	218	}
	219	#endif
	220	}
	221	}
	222
	223	/* Perform the DCT */
	224	(*do_dct) (workspace);
	225
	226	/* Quantize/descale the coefficients, and store into coef_blocks[] */
	227	{ register DCTELEM temp, qval;
	228	register int i;
	229	register JCOEFPTR output_ptr = coef_blocks[bi];
	230
	231	for (i = 0; i < DCTSIZE2; i++) {
	232	qval = divisors[i];
	233	temp = workspace[i];
	234	/* Divide the coefficient value by qval, ensuring proper rounding.
	235	* Since C does not specify the direction of rounding for negative
	236	* quotients, we have to force the dividend positive for portability.
	237	*
	238	* In most files, at least half of the output values will be zero
	239	* (at default quantization settings, more like three-quarters...)
	240	* so we should ensure that this case is fast. On many machines,
	241	* a comparison is enough cheaper than a divide to make a special test
	242	* a win. Since both inputs will be nonnegative, we need only test
	243	* for a < b to discover whether a/b is 0.
	244	* If your machine's division is fast enough, define FAST_DIVIDE.
	245	*/
	246	#ifdef FAST_DIVIDE
	247	#define DIVIDE_BY(a,b) a /= b
	248	#else
	249	#define DIVIDE_BY(a,b) if (a >= b) a /= b; else a = 0
	250	#endif
	251	if (temp < 0) {
	252	temp = -temp;
	253	temp += qval>>1; /* for rounding */
	254	DIVIDE_BY(temp, qval);
	255	temp = -temp;
	256	} else {
	257	temp += qval>>1; /* for rounding */
	258	DIVIDE_BY(temp, qval);
	259	}
	260	output_ptr[i] = (JCOEF) temp;
	261	}
	262	}
	263	}
	264	}
	265
	266
	267	#ifdef DCT_FLOAT_SUPPORTED
	268
	269	METHODDEF(void)
	270	forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
	271	JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
	272	JDIMENSION start_row, JDIMENSION start_col,
	273	JDIMENSION num_blocks)
	274	/* This version is used for floating-point DCT implementations. */
	275	{
	276	/* This routine is heavily used, so it's worth coding it tightly. */
	277	my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
	278	float_DCT_method_ptr do_dct = fdct->do_float_dct;
	279	FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
	280	FAST_FLOAT workspace[DCTSIZE2]; /* work area for FDCT subroutine */
	281	JDIMENSION bi;
	282
	283	sample_data += start_row; /* fold in the vertical offset once */
	284
	285	for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
	286	/* Load data into workspace, applying unsigned->signed conversion */
	287	{ register FAST_FLOAT *workspaceptr;
	288	register JSAMPROW elemptr;
	289	register int elemr;
	290
	291	workspaceptr = workspace;
	292	for (elemr = 0; elemr < DCTSIZE; elemr++) {
	293	elemptr = sample_data[elemr] + start_col;
	294	#if DCTSIZE == 8 /* unroll the inner loop */
	295	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	296	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	297	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	298	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	299	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	300	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	301	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	302	workspaceptr++ = (FAST_FLOAT)(GETJSAMPLE(elemptr++) - CENTERJSAMPLE);
	303	#else
	304	{ register int elemc;
	305	for (elemc = DCTSIZE; elemc > 0; elemc--) {
	306	*workspaceptr++ = (FAST_FLOAT)
	307	(GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
	308	}
	309	}
	310	#endif
	311	}
	312	}
	313
	314	/* Perform the DCT */
	315	(*do_dct) (workspace);
	316
	317	/* Quantize/descale the coefficients, and store into coef_blocks[] */
	318	{ register FAST_FLOAT temp;
	319	register int i;
	320	register JCOEFPTR output_ptr = coef_blocks[bi];
	321
	322	for (i = 0; i < DCTSIZE2; i++) {
	323	/* Apply the quantization and scaling factor */
	324	temp = workspace[i] * divisors[i];
	325	/* Round to nearest integer.
	326	* Since C does not specify the direction of rounding for negative
	327	* quotients, we have to force the dividend positive for portability.
	328	* The maximum coefficient size is +-16K (for 12-bit data), so this
	329	* code should work for either 16-bit or 32-bit ints.
	330	*/
	331	output_ptr[i] = (JCOEF) ((int) (temp + (FAST_FLOAT) 16384.5) - 16384);
	332	}
	333	}
	334	}
	335	}
	336
	337	#endif /* DCT_FLOAT_SUPPORTED */
	338
	339
	340	/*
	341	* Initialize FDCT manager.
	342	*/
	343
	344	GLOBAL(void)
	345	jinit_forward_dct (j_compress_ptr cinfo)
	346	{
	347	my_fdct_ptr fdct;
	348	int i;
	349
	350	fdct = (my_fdct_ptr)
	351	(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
	352	SIZEOF(my_fdct_controller));
	353	cinfo->fdct = (struct jpeg_forward_dct *) fdct;
	354	fdct->pub.start_pass = start_pass_fdctmgr;
	355
	356	switch (cinfo->dct_method) {
	357	#ifdef DCT_ISLOW_SUPPORTED
	358	case JDCT_ISLOW:
	359	fdct->pub.forward_DCT = forward_DCT;
	360	fdct->do_dct = jpeg_fdct_islow;
	361	break;
	362	#endif
	363	#ifdef DCT_IFAST_SUPPORTED
	364	case JDCT_IFAST:
	365	fdct->pub.forward_DCT = forward_DCT;
	366	fdct->do_dct = jpeg_fdct_ifast;
	367	break;
	368	#endif
	369	#ifdef DCT_FLOAT_SUPPORTED
	370	case JDCT_FLOAT:
	371	fdct->pub.forward_DCT = forward_DCT_float;
	372	fdct->do_float_dct = jpeg_fdct_float;
	373	break;
	374	#endif
	375	default:
	376	ERREXIT(cinfo, JERR_NOT_COMPILED);
	377	break;
	378	}
	379
	380	/* Mark divisor tables unallocated */
	381	for (i = 0; i < NUM_QUANT_TBLS; i++) {
	382	fdct->divisors[i] = NULL;
	383	#ifdef DCT_FLOAT_SUPPORTED
	384	fdct->float_divisors[i] = NULL;
	385	#endif
	386	}
	387	}