jcdctmgr.c

这是在PCA下的基于IPP库示例代码例子,在网上下了IPP的库之后,设置相关参数就可以编译该代码.

        *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;
      }
    }
  }
}


METHODDEF(void)
forward_DCT_intellib(
  j_compress_ptr       cinfo,
  jpeg_component_info* compptr,
  JSAMPARRAY           sample_data,
  JBLOCKROW            coef_blocks,
  JDIMENSION           start_row,
  JDIMENSION           start_col,
  JDIMENSION           num_blocks)
{
  /* This routine is heavily used, so it's worth coding it tightly. */
  JDIMENSION  bi;
  my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
  DCTELEM*    divisors = fdct->divisors[compptr->quant_tbl_no];
  Ipp8u       workspace[DCTSIZE2];  /* work area for FDCT subroutine */
  Ipp8u*      workspaceptr;

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

  for(bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE)
  {
    /* Load data into workspace */
    register JSAMPROW elemptr;
    register int      elemr;
    register JCOEFPTR output_ptr = coef_blocks[bi];

    workspaceptr = workspace;

    for(elemr = 0; elemr < DCTSIZE; elemr++)
    {
      elemptr = sample_data[elemr] + start_col;
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
      *workspaceptr++ = GETJSAMPLE(*elemptr++);
    }

    ippiDCTQuantFwd8x8LS_JPEG_8u16s_C1R(workspace,8,output_ptr,(Ipp16u*)divisors);
  }

  return;
} /* forward_DCT_intellib() */


#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:
    if(cinfo->UseIPP) {
      fdct->pub.forward_DCT = forward_DCT_intellib;
    } else {
      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
  }
}