jcdctmgr.c~

来自「JPEG Image compression using IJG standar」· C~ 代码 · 共 674 行 · 第 1/2 页

/*
 * 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;

  int i, j; /* JIE */
  JCOEFPTR tmp0;

  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);

/*******************/
/* Jie: test code */
/*******************/  
	/*
fprintf(stderr, "\nQuantization Table:\n");
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10d", divisors[i*8 + j]);
	}
	fprintf(stderr, "\n");
	}
	*/

	if (first_blk == 1){
  fprintf(stderr, "\nAfter forward DCT:(This is trueDCT * (Scale * 8)\n");
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10.2f", (float) workspace[i*8 + j]);
	}
	fprintf(stderr, "\n");
  }
	}
	first_blk = 0;
	
    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    { register DCTELEM temp, qval;
      register int i;
      register JCOEFPTR output_ptr = coef_blocks[bi];

	  if (lossless_codec) {

		/* skip quantization if lossless is specified. */
		for (i = 0; i < DCTSIZE2; i++) {
		  output_ptr[i] = (JCOEF) workspace[i];
		}
	
	  } else {

      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;
      }
    }

  } 
	

/*******************/
/* Jie: test code */
/*******************/
	/*
  fprintf(stderr, "\nAfter quantization: trueDCT/Q0)\n");
  tmp0 = coef_blocks[0];
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10d", tmp0[i*8 + j]);
	}
	fprintf(stderr, "\n");
  }
	*/

  }
}


#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. */
{
  int i, j;/* remove */
  JCOEFPTR tmp0;


  /* 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);

/*******************/
/* Jie: test code */
/*******************/  
	/*
fprintf(stderr, "\nQuantization Table:\n");
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10.2f", 1 / divisors[i*8 + j]);
	}
	fprintf(stderr, "\n");
  }

  fprintf(stderr, "\nAfter forward DCT:(This is trueDCT * (Scale * 8)\n");
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10.2f", (float) workspace[i*8 + j]);
	}
	fprintf(stderr, "\n");
  }
	*/

    /* 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);
      }
    }

/*******************/
/* Jie: test code */
/*******************/
	/*
  fprintf(stderr, "\nAfter quantization:(With Qfactor=50: divide by Q0 and 8Scale: get trueDCT/Q0)\n");
  tmp0 = coef_blocks[0];
  for (i = 0; i<8; i++){
	for (j = 0; j < 8; j ++) {
	  fprintf(stderr, "%10d", tmp0[i*8 + j]);
	}
	fprintf(stderr, "\n");
  }
	*/
  }
}

#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

	/* Jie 05/18/00*/
#ifdef DCT_BIN_A1_SUPPORTED
  case JDCT_BIN_A1:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_bin_a1;
    break;
#endif
#ifdef DCT_BIN_B1_SUPPORTED
  case JDCT_BIN_B1:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_bin_b1;
    break;
#endif
#ifdef DCT_BIN_C1_SUPPORTED
  case JDCT_BIN_C1:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_bin_c1;
    break;
#endif
#ifdef DCT_BIN_L1_SUPPORTED
  case JDCT_BIN_L1:
    fdct->pub.forward_DCT = forward_DCT;
    fdct->do_dct = jpeg_fdct_bin_l1;
    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
  }
}