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[compptr->component_index];
DCTELEM *divisors = ( DCTELEM * ) compptr->dct_table;
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 += compptr->DCT_h_scaled_size )
{
/* Perform the DCT */
( *do_dct )( workspace, sample_data, start_col );
/* 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;
}
}
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. */
j_lossy_c_ptr lossyc = (j_lossy_c_ptr) cinfo->codec;
fdct_ptr fdct = (fdct_ptr) lossyc->fdct_private;
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 comDicomon 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;
}
}
LOCAL void
extract_block (JSAMPARRAY input_data, int start_row, long start_col,
JBLOCK output_data, QUANT_TBL_PTR quanttbl)
/* Extract one 8x8 block from the specified location in the sample array; */
/* perform forward DCT, quantization scaling, and zigzag reordering on it. */
{
/* This routine is heavily used, so it's worth coding it tightly. */
DCTBLOCK block;
#ifdef DCT_ERR_STATS
DCTBLOCK svblock; /* saves input data for comparison */
#endif
{ register JSAMPROW elemptr;
register DCTELEM *localblkptr = block;
register int elemr;
for (elemr = DCTSIZE; elemr > 0; elemr--) {
elemptr = input_data[start_row++] + start_col;
#if DCTSIZE == 8 /* unroll the inner loop */
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
#else
{ register int elemc;
for (elemc = DCTSIZE; elemc > 0; elemc--) {
*localblkptr++ = (DCTELEM) (GETJSAMPLE(*elemptr++) - CENTERJSAMPLE);
}
}
#endif
}
}
#ifdef DCT_ERR_STATS
MEMCOPY(svblock, block, SIZEOF(DCTBLOCK));
#endif
j_fwd_dct(block);
{ register JCOEF temp;
register QUANT_VAL qval;
register int i;
for (i = 0; i < DCTSIZE2; i++) {
qval = *quanttbl++;
temp = (JCOEF) block[ZAG[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) (a >= b) ? (a /= b) : (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_data++ = temp;
}